Travel Demand Modeling

This note provides an overview of travel demand modeling.

The overall methodology for application of a modern transport model typically consists of:

  • Establishing a basis for categorizing (dividing) travel demand by passenger and cargo travel and by trip purpose and cargo type;
  • Recognizing the key underlying socio-economic variables that are the causes (generators) of travel demand;
  • Identifying the types and sources of data that will be needed for systematic analysis of the existing travel demand and for projection of future travel demand.
  • Collecting, processing and tabulating such data;
  • Analyzing the relationships of travel demand to its underlying causes;
  • Developing and calibrating traffic projection methods and models; and
  • Completing traffic projections for alternative scenarios and transport network expansion and improvement concepts.

Using information obtained from existing sources together with the information collected from surveys, the traffic/socio-economic data is analyzed to determine how to predict the essential characteristics of travel demand and travel patterns in the study area.  With respect to the investigation of travel demand, these analyses seek to determine the essential factors and relationships affecting the major components of travel behavior.

Travel demand modeling entails a number of steps and processes.  The most important of these are described briefly in the following sections.  Details of these and other additional areas can be found in any standard transportation planning book such as one by J. de D. Ortúzar and L.G. Willumsen (Modelling Transport, John Wiley and Sons, Second Edition 1995.)

Trip purposes

Because travel behavior depends on trip purpose (travel to work, education, shopping or other trip purpose site), an initial step in design of a travel demand modeling process will be the selection of the number and types of trip purposes to be separately modeled.  Typically, the trip purposes of personal travel which are separately modeled are: home based – work trips, home based – education trips, home based – other purpose trips, and non-home based trips.

Another consideration in the analysis of travel demand is that the total trips made within the study area can include:

  • Trips with both trip ends (trip origin and trip destination) within the limits of the study area and trips having one or both trip ends located outside the limits of the study area; and
  • Trips made by study area residents and trips made by visitors to the study area.

Trip generation analyses

For each trip purpose, the trip generation analyses will be to determine the best method for predicting the production and attraction of trips for given zonal land-use and population characteristics.  The methods for determining trip generation rates to be selected as the best methods may be obtained from either: (a) regression analysis or (b) cross-classification techniques.  Trip generation rates could be developed from regression analysis of the correlation of trip productions within zones to the zones’ average household population and socio-economic characteristics.  Another set of regressions are done to correlate the trip attractions to zones’ total number of jobs by type or other zonal variables that are logically and statistically related to trip attractions.

The cross-classification technique, which provides a preferable (disaggregated) method for predicting trip generation, involves the stratification of all households within zones into household groups having similar characteristics and developing trip generation rates for each household group.  For the (non-home) attraction end of personal trips, the consultant will investigate and, if practical and appropriate, develop and apply trip attraction rates related to specific types and intensities of land use.  For example, trip attraction rates could be related to such land-use measures as square meters of land-use, square meters of floor space, number of employees, or other measures of land-use that can be logically and statistically related to trip generation.

Trip distribution analysis

Two basic types of trip distribution models are usually applied to predict the future distribution (origins and destinations) of trips: (a) a gravity model or (b) an average growth factor method.  The gravity model is applicable in cases where data from a home interview survey is available for calibration of such a model.  In the absence of home interview survey data, an average growth factor method (such as the Fratar model) can be used to project an existing base year trip table to a future year by applying traffic growth factors (for each traffic analysis zone) for the production and attraction of trips.

Modal choice analysis

Mode choice analysis concerns with the relative usage of the available alternative modes of travel by the person making trips for the different trip purposes.  Alternative modes of travel that can be included in the modal choice analysis are: private passenger car driver, motorcycle driver, riding with someone else in passenger car or motorcycle, metro/LRT, public bus passenger by type of bus (large, mini-bus), private bus passenger, taxi passenger, walking, and other modes of travel.

The Consultant should explain and discuss the pros and cons for addressing modal choice for the different trip purposes and at alternative stages within the overall modeling process.  Mode choice procedures could be applied: (1) within the trip generation model, (2) before or after trip distribution, or (3) possibly within the traffic assignment process.

The mode choice models used in practice are generally based on the logit formulations.  Logit models have gained widespread acceptability in modeling travel choices as faced by travelers in the region.  Such models takes as inputs the level of service characteristics (such as travel time, travel cost etc.) and calculates the respective shares of each of the competing modes (i.e., auto, bus, light rail, etc.)

The simplest form of the logit model can be expressed as:

Sharek = exp(Uk) / ∑iexp(Ui)

where:

  • Uk is the utility of option k, computed as ∑jbjXjk
  • bj is the estimated parameter for the explanatory variable j
  • Xjk is the value of explanatory variable j for option k.

More advanced forms of logit models include the nested logit formulations.  The nested logit models address some of the shortcomings of the simple multinomial logit models and are theoretically superior.  The nested logit is a generalized logit model where the groups of modes (nests) with similar characteristics perceive the competition differently than the modes within the same group.  Nested logit models allow grouping of similar options in hierarchies or nests.  With these models it is possible to represent the intermodal competition much better and the groupings of alternatives indicate the cross-elasticity among the alternatives.  Alternatives in a common nest show the same degree of increased sensitivity compared to alternatives not in the nest.

Traffic assignment

Essential features for the traffic assignment process  include:

  • Capabilities for network link assignments for Annual Average Daily Traffic (AADT) volumes as well as peak hour and by direction of travel,
  • Assignment of person trips as well as vehicle and equivalent Passenger Car Unit (PCU) trips,
  • Intersection turning movements, delays, queue lengths, etc.,
  • Alternative assignments techniques including all-or-nothing, capacity restrained assignments, equilibrium, multi-path assignments, etc.; and
  • Transit assignments.

The determination of zone-to-zone network travel impedance (usually travel time) is an important feature of any of the available traffic assignment techniques.  Within such traffic assignment models, the determination of zone-to-zone travel times requires the ability to simulate the operation of all network links and their junctions under the possible range of traffic loading to which they can be subjected.  Within traffic assignment model, this is accomplished by the use of speed-volume curves (one for each unique link or intersection design class) which predict the operating speed associated with traffic loading expressed in volume-capacity ratio.  For this reason, the appropriate design class for each link and junction in the highway network must be determined and, in the road network inventories, speed-volume curves defined for each of the link and junction classes.  This work will involve sophisticated use of the road inventory information and analysis of the travel time and delay survey results.

Calibration of the traffic assignment model will involve assignment of the base year trip tables to network describing the existing study area transport network.  Then a comparison is made to determine how closely the assignment results match the actual traffic at pre-determined screen and/or cordon lines.  The hierarchical priority for checking the model will be to compare the model’s predictions for total vehicle-kilometers of network travel; vehicles per day crossing screen line and cordon locations; vehicle-kilometers of travel in major corridors; and finally, individual network links.

Model Calibration and Validation

The transport modeling system will be calibrated so that the measures of travel demand and traffic flow (i.e., AADT, seasonal variation, daily variation, morning and evening flows) are accurately predicted for the base year situation with respect to land use, transport network characteristics and travel demand.

After completing the analyses and calibrating each significant component of the transport model, all components will be integrated into an overall operational model.  This overall model will then be run through all stages of modeling process in order to check how well the model is capable of replicating the existing network travel and operating characteristics.  This process of model calibration includes:

  • Making consistency checks of principal components of the model to ensure that they are capable of producing accurate estimates of the current traffic levels and patterns and network performance measures for the study area (accomplished by comparing the traffic predictions of the model with the actual observed base year traffic usage of the highway/public transit network);
  • Applying forecasts of socio-economic and land-use patterns to the model in order to have the model produce forecasts of future network travel; and
  • Using the completed and calibrated model to make assessments of traffic and transportation impacts resulting from the range of possible future development scenarios for the study area.

After assuring that all elements of the model are properly calibrated, the model will then be applied to the task of predicting how the transport network will perform (using these same measures of transport network operation) for the base year.  This exercise, known as model validation, involves comparing the model output with observed data at screen-lines and cordon-lines; comparison of traffic volumes; and statistical analysis of assigned versus counted link volumes by volume group.

Land-use and Socio-economic Forecasts

The most important inputs to the models for forecasting traffic demand are the future land use and socio-economic patterns in the study area.  It is therefore essential that the land-use and socio-economic forecasts be done using standard and reliable econometric methodologies.  The land-use and socio-economic variables that are required for trip-generation models include:

  • Type of activity (residential, commercial, industrial, etc.)
  • Density of activity (residential or commercial space per square km, number of employees per sq. m. of built space, etc.)
  • Population
  • Employment by type
  • Households
  • Household size
  • Automobile ownership

The data required for such forecasts are generally gathered from different government agencies (such as the Census Bureau, housing and urban development agencies, etc.).

Traffic Forecasts

Traffic forecasts are prepared using the traffic modeling system and the forecasts of socio-economic variables along with the likely conditions on the transport network of the study area.