National Academies Press: OpenBook

Travel Demand Forecasting: Parameters and Techniques (2012)

Chapter: Chapter 6 - Emerging Modeling Practices

« Previous: Chapter 5 - Model Validation and Reasonableness Checking
Page 89
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 89
Page 90
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 90
Page 91
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 91
Page 92
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 92
Page 93
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 93
Page 94
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 94
Page 95
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 95
Page 96
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 96
Page 97
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 97
Page 98
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 98
Page 99
Suggested Citation:"Chapter 6 - Emerging Modeling Practices." National Academies of Sciences, Engineering, and Medicine. 2012. Travel Demand Forecasting: Parameters and Techniques. Washington, DC: The National Academies Press. doi: 10.17226/14665.
×
Page 99

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.

89 Over the past few decades, because of escalating capital costs of new infrastructure and increasing concerns regard- ing traffic congestion, energy dependence, greenhouse gas emissions, and air quality, the originally supply-oriented focus of transportation planning has expanded to include the objective of addressing accessibility needs and problems by managing travel demand within the available transportation supply. Consequently, there has been an increasing interest in travel demand management strategies, such as mixed land use development, parking pricing, and congestion pricing, all of which attempt to change land use and transportation service characteristics to influence individual travel behavior and control aggregate travel demand. The evaluation of such demand management strategies using travel demand mod- els places more emphasis on the realistic representation of behavior to accurately reflect traveler responses to manage- ment policies. This realization has led to the consideration of the follow- ing issues, all of which have the potential to improve upon travel demand forecasts and enable more informed policy making: • Time-space constraints and interactions in the activity- travel decisions of an individual; • The accommodation of interindividual interactions in activity-travel decision making across individuals (such as joint participation in activities and travel, serve passenger trips, and allocation of responsibilities among individuals in a household); • The recognition of the linkages across trips within the same “tour” (i.e., chain of trips beginning and ending at a same location) of an individual and across activities/tours of the individual over the day; and • The explicit consideration of time as an all-encompassing continuous entity within which individuals make activity/ travel participation decisions. The result has been the increasing consideration of a fundamental behavioral paradigm referred to as an activity- based approach to travel demand modeling. TRB Special Report 288: Metropolitan Travel Forecasting— Current Practice and Future Direction (SR 288) is the product of a TRB study, funded by FHWA, FTA, and the Office of the Secretary of Transportation, to determine the national state of practice in metropolitan area travel demand forecasting and to recommend improvements (Committee for Deter- mination of the State of the Practice in Metropolitan Area Travel Forecasting, 2007). SR 288 recommends that the fed- eral government “support and provide funding for the con- tinued development, demonstration, and implementation of advanced modeling approaches, including activity-based models” and “continue support for the implementation of activity-based modeling and other advanced practices; con- siderably expand this support through deployment efforts in multiple urban areas.” Chapter 6 of SR 288 is devoted to advancing the state of the practice. The purpose of this chapter is to introduce the concepts of advanced modeling procedures such as activity-based mod- els, dynamic traffic assignment models, and traffic simulation models. It is not intended to provide comprehensive docu- mentation of these advanced models, but rather to describe how they work and how they differ from the conventional models discussed in the rest of the report. This discussion should not be construed as a recommen- dation that all urban areas should be planning to switch to these types of modeling approaches in the near future, nor should it be viewed as a statement that such advanced modeling approaches address all of the problems associated with conventional modeling approaches. However, with these advanced approaches becoming more prevalent, and the like- lihood that more areas will continue to switch to using them, it is desirable for the travel modeling community to become more familiar with them. C h a p t e r 6 Emerging Modeling Practices

90 6.1 The Activity-Based Approach The fundamental difference between the trip- and activity- based approaches is that the former approach directly focuses on “travel participation behavior” as the decision entity of interest, while the activity-based approach views travel as a demand derived from the need to pursue activities and focuses on “activity participation behavior.” The underlying phi- losophy of the activity-based approach is to better understand the behavioral basis for individual decisions regarding partic- ipation in activities in certain places at given times, and hence the resulting travel needs. This behavioral basis includes all the factors that influence the why, how, when, and where of performed activities and resulting travel. Among these factors are the needs, preferences, prejudices, and habits of individuals (and households), the cultural/social norms of the commu- nity, and the travel service characteristics of the surrounding environment. At a fundamental level, therefore, the activity-based approach emphasizes the point that the needs of the households are likely to be translated into a certain number of total activity stops by purpose followed by (or jointly with) decisions regarding how the stops are best organized. For example, consider a congestion pricing policy during the evening commute period along a corridor. Also, consider an individual who has the daily pattern shown in the top pattern of Figure 6.1, where the shopping stop during the evening commute is at a location that entails travel along the “to-be-priced” corri- dor (but assume that the person would not be traveling the “to-be-priced” corridor if she went directly home from work). In response to the pricing policy, the individual may now stop making the shopping stop during the evening commute but may generate another stop in the evening after returning home from work (see bottom pattern of Figure 6.1). If some of these post-home arrival stops are undertaken in the peak period, congestion may be simply transferred to other loca- tions in the network. The activity-based approach explicitly acknowledges the possibility of such temporal redistribu- tions in activity participation (and hence travel) by focusing on sequences or patterns of activity participation (using the whole day or longer periods of time as the unit of analysis), and thus is able to provide a holistic picture of policy effects. A second defining aspect of the activity-based approach is its use of “tours” as the basic element to represent and model travel patterns. Tours are chains of trips beginning and ending at a same location, say, home or work. The tour-based rep- resentation helps maintain consistency across, and capture the interdependency (and consistency) of the modeled choice attributes among, the activity episodes (and related travel characteristics) undertaken in the same tour. This approach contrasts with the trip-based approach that considers travel as a collection of “trips,” each trip being considered independent of other trips. The activity-based approach can lead to improved evalu- ations of the impact of policy actions because of the explicit consideration of the interrelationship in the choice attributes (such as time of participation, location of participation, and mode of travel) of different activity episodes within a tour and, therefore, the recognition of the temporal, spatial, and modal linkages among activity episodes within a tour. Take, for example, an individual who drives alone to work and makes a shopping stop on the way back home from work (see Figure 6.2). The home-work and work-home trips in this scenario are not independent. Now consider an improvement in transit between the home and the work place. The activity-based approach would recognize that the individual needs to make a stop on the return home from work and so may not predict a shift to transit for the work tour (including the home-work, work- shop, and shop-home trips), while a trip-based model would break the tour into three separate and independent trips— a home-based work trip, a nonhome-based nonwork trip, and a home-based nonwork trip—and would be more likely (and inaccurately so) to shift the morning home-based work trip contribution of the individual to transit. Home HomeWork Shop Home HomeWork HomeShop Figure 6.1. Temporal substitution of trips.

91 In fact, the close association between mode choice for the work commute and stop making along the way is now well established. For instance, a study of Austin area workers (Bhat, 2004) found that the drive-alone mode share was 70 percent for commuters who never stopped on the way to or from work, compared to 87 percent for commuters who sometimes made a stop. Correspondingly, the share of commuters who used transit or a nonmotorized mode was higher for indi- viduals who did not make a commute stop. A third defining feature of the activity-based approach relates to the way the time dimension of activities and travel is considered. In the trip-based approach, time is included as a “cost” of making a trip and a day is viewed as a combination of broadly defined peak and off-peak time periods (see, for exam- ple, the time-of-day modeling discussion in Section 4.9). On the other hand, the activity-based approach views indi- viduals’ activity-travel patterns as a result of their time use decisions within a continuous time domain. Individuals have 24 hours in a day (or multiples of 24 hours for longer periods of time) and decide how to use that time among (or allocate that time to) activities and travel (and with whom), subject to their sociodemographic, spatial, temporal, trans- portation system, and other contextual constraints. These decisions determine the generation and scheduling of trips. Hence, determining the impact of travel demand manage- ment policies on time use behavior is an important precursor step to assessing the impact of such policies on individual travel behavior. Take the example of a worker who typically leaves work at 5:00 p.m. (say, the start of the afternoon peak period), drives to a grocery 15 minutes away, spends about 25 minutes shop- ping, and then gets back home by 6:00 p.m. (Figure 6.3). In response to an early release from work policy designed by the employer that lets the employee off from work at 4:00 p.m. instead of 5:00 p.m., a naïve model system may predict that the person would be off the road and back home by 5:00 p.m. (i.e., before the peak period begins; see the middle pattern in Figure 6.3). But the individual, now released from work earlier and having more time on his hands after work, may decide to drive a longer distance to a preferred grocery where he spends more time shopping (70 minutes rather than 25 minutes) and may eventually return home only at 6:00 p.m. (see the bottom pattern of Figure 6.3). So, in the case of this individual, not only would the policy be ineffective in keep- ing the person off the road during the peak period, but also the longer time spent at the grocery (in emissions analysis terms, the “soak duration,” the period between successive trips when the vehicle is not operational) would have adverse air quality implications. The activity-based model is able to consider such interactions in space and time due to its emphasis on time use and thus can produce more informed evaluations of policy actions. Another feature of the activity-based approach is the rec- ognition of interactions among household members, which leads to the accommodation of linkages among trips of house- hold members. As a result, policy actions could have complex Figure 6.2. Trip sequencing and interrelationship in attributes of linked trips. Home Shopping Drive alone Drive alone Drive alone Transit Improvements Figure 6.3. Duration and timing of activities and trips. 5:15 pm 5:40 pm 6:00 pm Work 4:00 pm 4:15 pm Shop 4:40 pm 5:00 pm 4:30 pm 5:40 pm Shop Home Soak duration = 25 min Soak duration = 70 min

92 responses, as shown in Figure 6.4. Consider that Person 1 (the worker) was originally dropping off the child at school in the mornings and picking up the child from school in the eve- nings, as part of the commute. Assume a pricing strategy on a corridor that connects the school location and the worker’s work location. Because of this pricing policy, the worker may not pursue the drop-off/pick-up tasks himself and has a sim- ple home-work-home pattern (top pattern of Figure 6.4). But now Person 2 (the nonworker) generates drop-off and pick- up trips, perhaps supplemented with shopping stops during his drop-off/pick-up trips. Such an explicit modeling of interindividual interactions and the resulting joint travel is particularly important to exam- ine the effects of occupancy-specific tolling strategies such as HOV and HOT lanes (Davidson et al., 2007). Another way that household linkages in activities can have an effect on responses to policies is through a reluctance to change the spatial and temporal attributes of joint activity episode par- ticipations. For instance, serve passenger trips (such as drop- ping off/picking up children from daycare/school or other extracurricular activities) and joint social/recreational out- of-home activities of household members may not be moved around much because of schedule constraints. Acknowledg- ing such joint interactions can, therefore, potentially lead to a more accurate evaluation of policy actions. A final important feature of activity-based approaches relates to the level of aggregation of decision makers used in the estimation and application of the models. In the trip-based approach, several aspects of travel (number of trips produced and attracted from each zone, trip interchanges, and mode split) are usually (though not always) estimated and/or applied at a relatively aggregate level of decision makers (such as at the spatial level of travel analysis zones). The activity-based models, on the other hand, have the ability to relatively easily accommodate virtually any number of decision factors related to the sociodemographic characteristics of the individuals who actually make the activity-travel choices. Using micro- simulation techniques, activity-based models predict the entire activity-travel patterns at the level of individuals (while recognizing temporal/spatial constraints across individuals of a household due to joint activity participations and serve passenger activities). Such a methodology ensures a realistic, consistent, and integral prediction of activity-travel patterns, which should lead to the better aggregate prediction of travel flows on the network in response to demographic changes or policy scenarios. Thus the activity-based models are well equipped to forecast the longer-term changes in travel demand in response to the changes in the sociodemographic compo- sition and the activity-travel environment of urban areas, as well as in response to land use and transportation policies. 6.2 Activity-Based Travel Model Systems in Practice 6.2.1 Overall Process for Activity-Based Model Systems The overall process used in the implementation of an activity-based model system comprises a sequence of three broad steps: 1. Population synthesis; 2. Long-term choice models; and 3. Activity-based travel models. Activity-based model systems require as inputs the infor- mation on each (and every) individual and household of the population of the study area, because the systems simulate the Figure 6.4. Resource sharing—linkages among trips of household members. Person 1 ShopDrop-off child Pick-up child Person 2 Work Home Home

93 activity-travel patterns of each individual in the study area. Such disaggregate-level sociodemographic inputs are gener- ated by synthesizing (i.e., simulating) the population of the study area. This synthesis is achieved by using zonal-level (or other levels of geography such as the block level or parcel level) forecasts of sociodemographic variables (such as household size, structure, and income) as controls for sampling house- holds using data from sources such as the ACS PUMS. At the end, the population synthesis procedure provides a synthetic sample of all households and individuals in the study area with information on household residential locations and all control variables used in the synthesis procedure. Several other socioeconomic attributes (which are not used as control variables) required by the activity-travel models are either directly borrowed from the households drawn from the PUMS data, or generated by a separate set of disaggre- gate models. The use of separate disaggregate models has the advantage that it provides natural variation in the predicted socioeconomic attributes, rather than “replicating” PUMS individuals and households. Some activity-based systems generate the synthetic population based on a two-way con- trol mechanism for both household-level attributes as well as individual-level attributes. After the population synthesis, the longer-term decisions such as auto ownership, work locations, and school locations are determined to recognize that such decisions are longer- term decisions that are not adjusted on a daily basis. Subse- quent to the determination of long-term choices, the synthetic population of households and individuals is “processed” through the activity-based travel model system, as discussed in more detail in the following sections. 6.2.2 Generic Structure of Activity-Based Systems Activity-based model systems used in practice typically consist of a series of utility maximization-based discrete choice models (i.e., multinomial logit and nested logit models) that are used to predict several components of individuals’ activity- travel decisions. In addition to such utility maximization- based model components, some model systems employ other econometric structures, including hazard-based duration structures and ordered response structures to model vari- ous activity-travel decisions. In effect, these model systems employ econometric systems of equations to capture rela- tionships between individual-level sociodemographics and activity-travel environment attributes on the one hand and the observed activity-travel decision outcomes on the other. As of 2011, MPOs within the United States that have devel- oped an activity-based travel model include Portland, Ore- gon; San Francisco, Sacramento, and Los Angeles, California; New York, New York; Columbus, Ohio; Denver, Colorado; and Atlanta, Georgia. Several other urban areas have activity- based models under development. While there are quite substantial variations among the many activity-based modeling systems in the precise sequence and methods used to predict the entire activity-travel pattern of each individual, all of these systems essentially include a three-tier hierarchy of (1) day-level activity pattern choice models (or, simply, pattern-level choice models); (2) tour- level choice models; and (3) trip/stop-level choice models. The choice outcomes from models higher in the hierarchy (assumed to be of higher priority to the decision maker) are treated as known in the lower-level models. The pattern- level models typically provide a skeletal daily pattern for each individual, including whether the individual goes to work (or school, if the person is a student), whether the individual takes any children to/from school, any joint activities (and their purposes) among individuals in a household and the individuals involved, individual participations in activities by purpose, and number of total tours (home- and work-based) in the day. The tour-level models typically determine the number of stops in a tour by purpose and their sequence, the travel mode for the tour, and the time of day and duration of the tour. For workers, tours are constructed based on focusing on the home-work and work-home commutes first, along with the number of stops, sequence, and travel mode during the commutes. Next, other tours during the day are con- structed; those with joint activities are usually given sched- uling precedence. For nonworkers, tours relating to serve passenger stops (including dropping off/picking up children from school/day care) and tours with joint activities may get scheduling precedence. Finally, the stop-level models predict the stop location, mode choice, and time of day of travel for each of the stops in each tour. 6.2.3 Data Needs for Estimation of Activity-Based Systems The primary sources of data for the estimation of tour- and activity-based models are household activity and/or travel surveys. As the term “household activity and/or travel sur- veys” suggests, the surveys can be either travel surveys (that collect information on out-of-home travel undertaken by the household members) or activity-travel surveys (that collect information on out-of-home activities and associated travel). Both the surveys implicitly or explicitly collect information on (1) household-level characteristics, (2) individual-level characteristics, and (3) information on the activity/travel epi- sodes undertaken by the individuals. Activity surveys, how- ever, also may collect additional information on individuals’ activities, specifically the participation in, timing, and dura- tion of in-home and joint activities.

94 It should be noted that the development of several activity- based models to date has involved the use of household travel survey data that are not any different from those collected and used by regional MPOs for their trip-based model development and calibration. Thus, the notion that activity-based models are data hungry is not necessarily accurate, at least at the esti- mation stage (though, activity-based models would perhaps benefit more from larger sample sizes than would trip-based models, especially from the standpoint of estimating models of joint activity participation). The estimation of activity- based models does require more extensive efforts (relative to a trip-based approach) in preparing the data to construct the entire sequence of activities and travel, but such intense scru- tiny of data also helps identify data inconsistencies that might go unchecked in the trip-based approach. For example, there might be “gaps” in an individual’s travel diary because of non- reporting of several trips; these will be identified during data preparation for activity analysis but may not be identified in the trip-based approach because it highlights individual trips and not the sequence between trips and activities. Data on regional land use and transportation system net- works also are typically used in model estimation. Land use data include information on the spatial residential charac- teristics of households, employment locations, and school and other locations at the level of spatial resolution (for exam- ple, zones or parcels) used in the models. The typical land use information includes size and density measures, such as number of households, population, area (or size), employ- ment by each category of employment, household density, population density, and employment density for each cat- egory of employment. In addition, one or more of the fol- lowing land use data also are used by some activity modeling systems: (1) land use structure information, such as the per- centage of commercial, residential, other developed, and open areas; percentage of water coverage; and the land use mix; (2) sociodemographic characteristics, such as average house- hold size, median household income, ethnic composition, housing characteristics such as median housing value, and housing type measures (single- and multiple-family dwelling units); and (3) activity opportunity measures such as activity center intensity (i.e., the number of business establishments within a fixed network distance) and density (i.e., the num- ber of business establishments per square mile) for each of several activity purposes. Transportation network data needed in activity models are similar to data used in trip-based models and typically include highway network data, transit network data, and nonmotorized mode data. The transportation system perfor- mance data should be of high quality, with time-varying LOS characteristics (in-vehicle, out-of-vehicle, access, egress, and wait times) across different time periods, as well as across different location pairs. 6.2.4 Data Needs for Application of Activity-Based Systems Once the activity-based modeling system has been esti- mated using the data sources discussed in the previous sec- tion, the application of these activity-based models for a study area for a base year requires as inputs the information on all individuals and households of the study area for the base year. Synthetic population generation techniques are used for this purpose, sometimes supplemented with a series of other demographic models (see Section 6.2.1). For a future-year forecasting exercise, the inputs should consist of the future- year synthetic population and land use and LOS data. Thus, activity-based model development should be supported with the development of detailed input data (i.e., the synthetic population and LOS and land use data) for future years. This can be done either by using aggregate demographic and land use projections for future years and applying a synthetic pop- ulation generator (just as in the base year) or “evolving” the base-year synthetic population (see Eluru et al., 2008). More details on this are provided in Section 6.3.1. 6.2.5 Data Needs for Calibration and Validation of Activity-Based Systems The following data sources can be used to calibrate and validate activity-based model systems: • Validation of input data – The base-year synthetic population inputs can be vali- dated against census data. – To validate the input work locations, the home-work trip lengths and patterns can be matched against those in observed data sources such as CTPP. – To validate the vehicle ownership inputs, census data and perhaps other sources such as motor vehicle depart- ment estimates of auto registrations can be used. • Calibration and validation of activity-travel outputs – Each component of the activity-travel model system can be validated by comparing its predictions to the observed activity-travel patterns in the household activity-travel survey. – The commute mode choice model can be validated using data such as CTPP. – The entire model system can be validated by comparing the traffic assignment outputs with the observed traffic volumes in the study area. – Highway traffic assignment validation can be undertaken by using observed traffic volumes by time of day, while transit traffic assignment validation can be pursued by using transit boarding/alighting data by route and stop by time of day from an on-board transit survey/count.

95 Along with the above-identified base-year calibrations and validations, it is essential to understand the forecasting ability and the policy sensitivity of activity-based models for nonbase-year conditions. To test the forecasting ability, the model performance for past years (for example, year 1990) and for existing “future” years relative to the base year for the travel modeling effort (for example, year 2010) can be compared with the observed patterns in those years. For this purpose, complete input data (including the aggregate sociodemographic variable distribu- tions for synthetic population generation, and the land use and LOS data), observed traffic volumes, household activity- travel survey data, and the census data (if available) are required for past years and existing “future” years. In this regard, it is important that the regional planning agencies store and document the land use data and transportation network data of past and existing “future” years. An examination of the policy sensitivity of activity-based models for nonbase-year conditions can be undertaken by assessing the impact on activity-travel patterns of changes in transportation system and land use patterns. To this end, in the recent past, several tests have been undertaken to assess the sensitivities of specific components of activity-based mod- els to policy scenarios. Examples include (1) an analysis of the impact of LOS changes (systemwide and localized); (2) analy- ses of capacity expansion and centralized employment sce- narios; (3) analysis of area pricing schemes; (4) assessment of the effect of shortened work days; (5) analyses of cordon pricing and increased transportation network connectivity scenarios; (6) user-benefit forecasts of light rail transit proj- ects; (7) equity analysis of transportation investment impacts; (8) examination of the impacts of land use and urban form on area travel patterns; (9) analysis of congestion pricing policies; (10) analysis of FTA New Starts projects; and (11) analysis of transit investments. Such an examination of the response to several policy scenarios can be a useful assessment of the abilities of the activity-based model system (especially when compared with the outputs from a trip-based model system). The scenario approach discussed above to assessing the policy sensitivity of activity-based models, however, may not completely represent the complexity of real-life projects and policies. Furthermore, sensitivity testing using test scenarios serves only as a broad qualitative reasonableness assessment of performance, rather than a quantitative performance measure- ment against observed data. A more robust way to quantify and assess the predicted policy sensitivity from activity-based models is to compare the model predictions with real-world data before—and after—real-life transportation infrastruc- ture investments or policy actions. Hence, it is important to collect traffic counts and other travel pattern data before— and after—any major transportation infrastructure invest- ments or policy actions. 6.2.6 Software for Activity-Based Modeling At present, there are no readily available standard software packages to apply activity-based models. The model systems developed for various MPOs have been developed and imple- mented as customized stand-alone software, and then integrated with standard proprietary modeling software for such purposes as network skimming, matrix manipulation, and highway and transit assignment. Most activity model systems are coded using C++, C#, Python, or Java and make use of an object-oriented approach, which offers the advantages of code reuse, software extensibility, and rapid implementation of system variants. 6.2.7 Challenges of Developing and Applying Activity-Based Modeling Systems The development of activity-based models requires careful and extensive data preparation procedures to construct entire “sequences” of activities and “tours” of travel. The data prepa- ration process for the activity-based modeling is involved and requires skilled and experienced personnel. Furthermore, as mentioned previously, activity-based model development is associated with an initial overhead of data preparation, model estimation, calibration and validation, and the process of “putting it all together” into customized software. How- ever, once the model system is developed, the system can be packaged as user-friendly travel demand modeling and pol- icy analysis software. Further, the software can be sufficiently generic to allow its use in any study area, provided the model parameters for that area are available. The implementation of activity-based models (for either the base year or for future years) requires the end user to be well aware of the details of the system. Another implementa- tion challenge is the significant amount of run time, because activity-based models simulate the activity-travel patterns of each (and every) individual of a study area. However, it appears that the run times can be significantly reduced by one or more of the following techniques: • Simulation of the activity-travel patterns of a sample of the population without substantially compromising the accu- racy of the aggregate-level outputs; • Efficient computing strategies such as data caching and multi-threading; • “Clever” methods of model specification where dummy exogenous variables are used so that a substantial part of the computations in the application context can be under- taken for market segments (defined by combinations of dummy exogenous variables) rather than for each indi- vidual in the population; and • Use of cloud (or cluster) computing approaches that use several parallel processors at the same time.

96 The implementation challenges associated with activity- based models appear to be higher for the forecast-year imple- mentation rather than for the base-year implementation, primarily because of the need to generate detailed socio- economic input data for the forecast years. Also, the develop- ment of future-year parcel-level land use data is a challenge associated with the implementation of models that use parcel- level data. And in rapidly growing areas, there may be many more synthetic persons and households to simulate than in the base year. Finally, while the required technical background, resource requirements for development and maintenance, implemen- tation challenges, and institutional issues associated with ownership of activity-based models are immediately evident, the need remains to assess, document, and demonstrate the potential practical benefits of these models. 6.3 Integration with Other Model Systems The recognition of the linkages among sociodemographics, land use, and transportation is important for realistic forecasts of travel demand, which has led practitioners and researchers to develop approaches that capture sociodemographic, land use, and travel behavior processes in an integrated manner. Such behavioral approaches emphasize the interactions among population socioeconomic processes; the households’ long- term choice behaviors; and the employment, housing, and transportation markets within which individuals and house- holds act (see Waddell, 2001). From an activity-travel forecast- ing perspective, these integrated urban modeling systems need to consider several important issues that are outlined in this section. Some elements of this integration with activity-based models already have been introduced at several MPOs. 6.3.1 Generation of Disaggregate Sociodemographic Inputs for Forecast Years As indicated in Section 6.2.3, activity-based travel forecast- ing systems require highly disaggregate sociodemographics as inputs, including data records of each and every individual and household in the study area. Hence, disaggregate popula- tion generation procedures are used to create synthetic records of each and every individual and household for activity- travel microsimulation purposes. However, to be able to forecast the individual activity-travel patterns and aggregate transportation demand at a future point in time, activity- based travel demand models require, as inputs, the disaggre- gate sociodemographics, and the land use and transportation system characteristics of that future point in time. While synthetic population generator (SPG) procedures can be used for this purpose as a first step operationalization strategy, these procedures work off aggregate demographic and land use projections for future years rather than the more desirable route of evolving the base-year population. Spe- cifically, individuals and households evolve through a socio- demographic process over time. As the sociodemographic process unfolds, individuals may move into or out of life-cycle stages such as schooling, the labor market, and different jobs. Similarly, households may decide to own a house as opposed to rent, move to another location, and acquire/dispose of a vehicle. Such sociodemographic processes need to be mod- eled explicitly to ensure that the distribution of population attributes (personal and household) and land use characteris- tics are representative at each point of time and are sufficiently detailed to support the activity-travel forecasting models. There have been relatively limited attempts to build mod- els of sociodemographic evolution for the purpose of travel forecasting. Examples in the transportation field include the CEMSELTS system by Bhat and colleagues (Eluru et al., 2008), the DEMOgraphic (Micro) Simulation (DEMOS) sys- tem by Sundararajan and Goulias (2003), and the Micro- analytic Integrated Demographic Accounting System (MIDAS) by Goulias and Kitamura (1996). Examples from the non- transportation field include DYNACAN (Morrison, 1998), and LIFEPATHS (Gribble, 2000). 6.3.2 Connecting Long- and Short-Term Choices Many (but not all) operational activity-based travel demand models treat the longer-term choices concerning the hous- ing (such as residential tenure, housing type, and residen- tial location), vehicle ownership, and employment choices (such as enter/exit labor market and employment type) as exogenous inputs. Consequently, the land use (in and around which the individuals live, work, and travel) is treated as exogenous. In such cases, the possibility that households can adjust with combinations of short- and long-term behavioral responses to land use and transportation policies is system- atically ignored (see Waddell, 2001). A significant increase in transportation costs, for example, could result in a household adapting with any combination of daily activity and travel pat- tern changes, vehicle ownership changes, job location changes, and residential location changes. While many travel forecasting models treat the long-term choices and hence the land use as exogenous to travel behav- ior, there have been recent attempts to model the longer- and shorter-term choices in an integrated manner. These include OPUS/UrbanSim (Waddell et al., 2006), CEMUS (Eluru et al., 2008), ILUTE (Salvini and Miller, 2005), and ILUMASS (Strauch et al., 2003). There also have been models studying the relationships between individual elements of land use- related choices and travel behavior choices. However, most

97 of these models and model systems are trip based. That is, although these models attempt to study the land use and travel behavior processes in an integrated manner, the travel behav- ior aspect of these models is based on a trip-based approach. 6.3.3 Demand-Supply Interactions The end use of travel forecasting models is, in general, the prediction of traffic flow conditions under alternative socio- demographic, land use, and transportation LOS scenarios. The traffic flow conditions, which are usually predicted after a traffic assignment procedure, are a result of the interactions between the individual-level demand for travel and the travel options and LOS (or the capacity) supplied by the transportation sys- tem. At the same time, the activity-travel patterns predicted by an activity-based modeling system (that are input into traffic assignment) are themselves based on specified LOS values. Thus, as in a traditional trip-based model, one needs to ensure that the LOS values obtained from the traffic assignment proce- dure are consistent with those used in the activity-based model for activity-travel pattern prediction. This is usually achieved through an iterative feedback process (see Section 1.3) between the traffic assignment stage that outputs link flows/LOS and the activity-based travel model that outputs activity-travel patterns. It is important to consider such demand-supply interactions for accurate predictions of activity-travel behavior, and the result- ing traffic flow conditions. Further, since the travel LOS varies with the temporal variation in travel demand, and the demand for travel is, in turn, dependent on the transportation level of service, the interactions may be time-dependent and dynamic in nature. Thus, it is important to consider the dynamics of the interactions between travel demand and the supply of transpor- tation capacity (see next section for additional details). Similar to how transportation market processes (i.e., the interactions between individual-level travel demand and the transportation supply) influence the individual-level activity- travel patterns, the housing and labor market processes influ- ence the residential and employment choices of individuals. In fact, individuals act within the context of, and interact with, housing, labor, and transportation markets to make their residential, employment, and activity-travel choices. While the transportation market process may occur over shorter timeframes (such as days or weeks), the employment and housing market processes are likely to occur over longer periods of time. That is, in the short term, the daily activity- travel patterns are directly influenced by the dynamics of the interaction between travel demand and supply; while in the long term, the activity-travel behavior is indirectly affected by the impact of housing and labor market processes on the residential and employment choices, and also on the land use and transportation system. If the activity-travel behavior of individuals and households is to be captured properly over a longer timeframe, the interactions with, and the evolution over time of, all these markets should be explicitly consid- ered, along with the sociodemographic processes and the long-term housing and employment choices. 6.3.4 Traffic Simulation The precise form of the interaction between an activity- based model and a traffic assignment model (as discussed in the previous section) depends on the nature of the assign- ment model used. In many places where activity-based mod- els have been implemented in practice, it is not uncommon to convert the activity-travel patterns into trip tables by travel mode for four to five broad time periods of the day, and then load the time period-specific trip tables using a traditional static traffic assignment (STA) methodology. This static assignment methodology uses analytic link volume-delay functions, combined with an embedded shortest path algo- rithm, to determine link flows and link travel times (see Sec- tion 4.11). In such a static assignment approach, there is, in general, no simulation of individual vehicles and no consid- eration of temporal dynamics of traffic flow. On the other hand, an important appeal of the activity-based approach is that it predicts activity-travel patterns at a fine reso- lution on the time scale. Thus, using an activity-based model with a static assignment process undoes, to some extent, the advantages of predicting activity-travel patterns at a fine time resolution. This limitation, and the increase in computing capacity, has allowed the field to move toward a dynamic traf- fic assignment (DTA) methodology. The DTA methodology offers a number of advantages relative to the STA methodol- ogy, including the ability to address traffic congestion, build- up, spillback, and oversaturated conditions through the explicit consideration of time-dependent flows and the representation of the traffic network at a high spatial resolution. As a result, DTA is able to capture and evaluate the effects of controls (such as ramp meters and traffic lights), roadway geometry, and intel- ligent transportation system (ITS) technology implementations. Some literature on analytical method-based DTA models exists. However, the implementation of most DTA models relies on a microsimulation platform that combines (and iterates between) a traffic simulation model (to simulate the movement of traffic) with time-dependent routing algo- rithms and path assignment (to determine flows on the network). In particular, the traffic simulation model takes a network (nodes, links, and controls) as well as the spatial path assignment as input, and outputs the spatial-temporal trajec- tories of vehicles as well as travel times. The time-dependent shortest path routing algorithms and path assignment models take the spatiotemporal vehicle trajectories and travel times as input, and output the spatial path assignment of vehicles. The two models are iterated until convergence between

98 network travel times and vehicle path assignments. In this process, the traffic simulation model used may be based on macroscopic traffic simulation (vehicle streams considered as the simulation entity and moved using link volume-delay functions), mesoscopic traffic simulation (groups of vehicles considered as “cells” and treated as the simulation entity), or microscopic traffic simulation (each individual vehicle con- sidered as the simulation entity, incorporating intervehicle interactions). Macroscopic and mesoscopic traffic simula- tion models are less data hungry and less computationally intensive than microscopic models, but also are limited in their ability to model driver behavior in response to advanced traffic information/management systems. Most earlier DTA efforts have focused on the modeling of private car traffic, though a few recent research efforts (see, for example, Rieser and Nagel, 2009) have integrated mode choice and departure time choice within a microsimulation- based DTA model, thus moving further upstream in integrat- ing activity-based models with dynamic traffic assignment. Recently, there have been other efforts under way that explore the complete integration of activity generation, scheduling, traffic simulation, route assignment, and network loading within a multiagent microsimulation platform. For exam- ple, Project C10 of the second Strategic Highway Research Program (SHRP 2), “Partnership to Develop an Integrated, Advanced Travel Demand Model,” is developing integrated models that include activity-based demand model and traffic simulation model components, taking advantage of the disaggregate application approach in both components (Cambridge Systematics, Inc. and National Academy of Sciences, 2009; Resource Systems Group and the National Academy of Sciences, 2010). Activity-based modeling also can be integrated with mod- els of transit passenger simulation. Person tours generated by the activity-based model that are fully or partially made via transit can have their transit paths simulated individually. This individual simulation requires the specification of all transit vehicle runs and stops and the assigning of passenger trips to these runs and stops, along with their walk and auto access and egress components. One of the SHRP2 C10 tasks is incorporating this capability. The greatest impediments to regionwide traffic simulation are the expensive computational resources and time needed (though distributed and parallel implementation designs are possible), and the costs and complexity of data acquisition/ management and model calibration (though GIS tools and GPS-based vehicle survey techniques are making this easier). Note that the use of DTA does not require an activity- based model; in fact, DTA has been used in connection with conventional (i.e., four-step) models for some time. In such cases, the aggregate results of the conventional models (i.e., trip tables) are converted to disaggregate lists of trips to be simulated. Thus, disaggregate activity-based demand models have often been used with aggregate assignment techniques, and aggregate demand models have been used with dis- aggregate assignment techniques. The connection between disaggregate demand and assignment models is the subject of much contemporary research and development. 6.3.5 Example of an Integrated Urban Modeling System In view of the preceding discussion, ideally, activity-based travel demand models should be integrated with other mod- els that can forecast, over a multiyear timeframe, the socio- demographic processes, the housing and employment market processes, and traffic flows and transportation system condi- tions. The integrated model system should be able to capture the previously discussed demand-supply interactions in the housing, employment, and transportation markets. A con- ceptual framework of such a system is provided in Figure 6.5. The integrated system places the focus on households and individuals, and businesses and developers that are the pri- mary decision makers in an urban system. The system takes as inputs the aggregate socioeconomics and the land use and transportation system characteristics for the base year, as well as policy actions being considered for future years. The aggregate-level base-year socioeconomic data are first fed into an SPG module to produce a disaggregate-level synthetic data set describing a subset of the socioeconomic characteristics of all the households and individuals residing in the study area. Additional base-year socioeconomic attributes—related to mobility, schooling, and employment at the individual level and residential/vehicle ownership choices at the household level—that are difficult to synthesize (or cannot be synthe- sized) directly from the aggregate socioeconomic data for the base year are simulated by the socioeconomics, land use, and transportation (SLT) system simulator. The base-year socioeconomic data, along with the land use and transportation system attributes, are then run through the daily activity-travel pattern (AT) simulator to obtain individual-level activity-travel patterns. The activity-travel patterns are subsequently passed through a dynamic traffic micro-assignment (DT) scheme to determine path flows, link flows, and transportation system level of service by time of day [see Lin et al. (2008) for a discussion of recent efforts on integrating an activity-travel simulator and a dynamic traffic microsimulator]. The resulting transportation system LOS characteristics are fed back to the SLT simulator to generate a revised set of activity-travel environment attributes, which is passed through the AT simulator along with the socio- economic data to generate revised individual activity-travel patterns. This “within-year” iteration is continued until base- year equilibrium is achieved. This completes the simulation for the base year. The next phase, which takes the population one step for- ward in time (i.e., 1 year), starts with the SLT simulator updat- ing the population, urban form, and the land use markets

99 Figure 6.5. An integrated model system. Source: Modified from Eluru et al. (2008). Activity-travel environment characteristics (base year) Subset of socioeconomic characteristics (base year) Activity-travel simulator (AT) Individual-level activity-travel patterns Path/link flows and level of service Dynamic Traffic Micro- Assignment (DT) Socioeconomics, land use and transportation system characteristics simulator (SLT) Socioeconomic characteristics and activity-travel environment Policy actions Aggregate socioeconomics (base year) Synthetic population generator (SPG) Base-Year Inputs Forecast-Year Outputs (note that SPG is used only to generate the disaggregate-level synthetic population for the base year and is not used beyond the base year). An initial set of transportation system attri- butes is generated by SLT for this next time step based on (1) the population, urban form, and land use markets for the next time step; (2) the transportation system attributes from the previous year in the simulation; and (3) the future-year policy scenarios provided as input to the integrated system. The SLT outputs are then input into the AT system, which interfaces with the DT scheme in a series of equilibrium iterations for the next time step (just as for the base year) to obtain the “one-time step” outputs. The loop continues for several time steps forward until the socioeconomics, land use, and transportation system path/link flows and transpor- tation system LOS are obtained for the forecast year speci- fied by the analyst. During this iterative process, the effects of the prescribed policy actions can be evaluated based on the simulated network flows and speeds for any intermediate year between the base year and the forecast year. 6.4 Summary Activity-based model systems are different from the con- ventional trip-based model systems in five major aspects. First, activity-based systems recognize that travel is derived from the need to pursue activities at different points in space and time, and thus focus on modeling activity participation. Second, activity-based model systems use a tour-based structure to rep- resent and model travel patterns. Tours are defined as chains of trips beginning and ending at a same location, say, home or work. Such representation captures the interdependency (and consistency) of the modeled choice attributes among the activity episodes of the same tour. Third, activity-based model systems view individuals’ activity-travel patterns as a result of their time use decisions within a continuous time domain, subject to their sociodemographic, spatial, tempo- ral, transportation system, and other contextual constraints. Fourth, activity-based systems accommodate for inter- actions and joint activity participations among individu- als in a household. Finally, activity-based systems simulate the activity-travel patterns of each (and every) individual of the study area using a microsimulation implementation that provides activity-travel outputs that look similar to survey data and can allow analysis of a wide range of poli- cies on specific sociodemographic segments. Activity-based travel models are increasingly being adopted by the larger MPOs in the country and offer a more compre- hensive and potentially more accurate assessment of policies to enhance mobility and reduce emissions. While the prin- ciple behind the activity-based analysis approach has existed for at least three decades now, it is only in the past 5 to 10 years that the approach has started to see actual imple- mentation. As a result, there has been no formal analysis of transferability of parameters and model structures in space and/or time in the context of activity-based models. This area will inevitably see increasing attention in the near future. Future versions of this report might include informa- tion on the potential transferability of activity-based mod- eling parameters and possibly some specific transferable parameters.

Next: Chapter 7 - Case Studies »
Travel Demand Forecasting: Parameters and Techniques Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s National Cooperative Highway Research Program (NCHRP) Report 716: Travel Demand Forecasting: Parameters and Techniques provides guidelines on travel demand forecasting procedures and their application for helping to solve common transportation problems.

The report presents a range of approaches that are designed to allow users to determine the level of detail and sophistication in selecting modeling and analysis techniques based on their situations. The report addresses techniques, optional use of default parameters, and includes references to other more sophisticated techniques.

Errata: Table C.4, Coefficients for Four U.S. Logit Vehicle Availability Models in the print and electronic versions of the publications of NCHRP Report 716 should be replaced with the revised Table C.4.

NCHRP Report 716 is an update to NCHRP Report 365: Travel Estimation Techniques for Urban Planning.

In January 2014 TRB released NCHRP Report 735: Long-Distance and Rural Travel Transferable Parameters for Statewide Travel Forecasting Models, which supplements NCHRP Report 716.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!