3. Model design

3.1 Modelling demand

3.1.1 Travel demands

Reference (Base) Year travel demands for highway, rail and ferry travel can be derived in a number of ways. The usual, and most expensive, way of collecting travel demand data is through travel surveys, either one-off or continuous. Survey methods include self-completion travel diaries, household interview surveys and in-vehicle public transport surveys. Vehicle counts are also useful in providing a database for the calibration as well as validation of the transport models.

Travel surveys can be structured to derive personal travel purpose origin–destination matrices for use in the assignment process by collecting information such as:

  • Origin and destination by purpose
  • Origin and destination by location
  • Car availability and use for travel purpose
  • Public transport mode used
  • Cost of travel
  • Duration of travel
  • Time of day of travel
  • Age, gender, income and employment status of the traveller.

3.1.2 Market segmentation

The starting point for analysis of travel demand is to note that travel is almost always a derived demand - that is, it only occurs because of some other underlying demand. Travel occurs and goods are shipped because people want to undertake specific activities at different locations.

Demand characteristics (price, income and cross elasticities, sensitivity to time, comfort for passengers, growth rates) and transport costs (type of service demanded) will vary for different segments of the market.

Person trips may be categorised in many ways, including factors such as trip purpose, trip frequency, trip timing, trip distance and spatial separation of origin and destination (O-D) as well as travel mode used. Further, the socio-economic characteristics of individual travellers and the households to which they belong are also important determinants in predicting the travel behaviour of those individuals. The breakdown shown in Table 2 can be considered broad brush categorisation of passenger travel.

Table 2: Classification of passenger travel in terms of trip purpose, trip frequency, trip timing, spatial separation and transport mode
PURPOSEFREQUENCYTRIP TIMINGTRAVEL DISTANCETRANSPORT MODE
Work

Education

Shopping

Personal business

Work related

Social

Recreational
Regular

Infrequent

Occasional

‘One-off’
Peak period

Business hours

Off-peak

Late night

Weekday

Weekend
Local

To city centre

Inter-suburb

Regional

Inter-city

Inter-state

International

Origin-destination*

within a locality separated, good access separated, difficult access through
Private car:
  • driver
  • passenger
Public transport
  • rail
  • bus
  • tram/LRT
  • taxi
  • ferry
  • aircraft
  • ship
Non-motorised
  • walk
  • bicycle

Greater modelling precision, and potentially improved model accuracy, can be obtained by a more refined market segmentation than displayed in Table 2. However, such refinement must be balanced against the data requirements to support its implementation.

Knowledge of spatial patterns of travel demand is used in transport planning for network and service design. A common method of describing travel demand in a region is through the use of origin-destination matrices. These are tables of trip or commodity movements between the various O-D pairs that exist in a study region (Taylor, Bonsall & Young 2000, pp. 114–116). Consider the schematic map of such a region as shown in Figure 4. The study region is identified by a cordon line around it. Travel movements across the cordon line indicate trips made to and from the region. These are external origins and destinations. Observations on the cordon line can be used to assess the numbers and patterns of these through trips. Internal or local trip movements may also exist for trips where either or both the origin and destination of the trip are located inside the study area. Further information on through trips (such as routes chosen for the segments of those trips inside the study area) and local trips can be gathered by defining screenlines inside the study area and then making observations of vehicle movements at the screenlines. Traffic management studies are often concerned with the proportion of through traffic to local traffic in the study area.

Figure 4: Origins and destinations and trip movements in an identified study region Origins and destinations and trip movements in an identified study region

Figure 4 shows different O-D configurations. These include through trips and local trips (those that have at least one trip end – origin or destination – inside the study area). Local trips may be further subdivided into categories of within locality, separated with good access and separated with difficult access, as indicated in the figure. A typical structure for an O-D table is given in Table 3. Travel movements may be expressed in units of vehicles, passengers or commodity flows.

Table 3: Typical structure of an origin-destination matrix of travel movements
ORIGINDESTINATION
Externa

lInternal
1

..

N1

1

..

N2
External

1 … M1

Internal

1 … M2

Through trips Local trips, destination in study region (separated trip, good / difficult access)
Local trips, origin in study region (separated trip, good / difficult access) Local trips (within locality)

Temporal distributions of travel demand are also important. Travel demand varies over the hours of the day, the days of the week and the weeks of the year. Cyclic and seasonal patterns can be ascertained to describe these patterns and used to predict the demands for transport services.

3.1.3 Peak periods

Specific time period (am peak, inter-peak, pm peak and off-peak) trip matrices should be developed to better reflect the different travel making propensities and characteristics during these periods. This approach is preferable in urban areas, where commuter peaks place the greatest loads on the available transport infrastructure.

Another approach is to develop daily travel demands and then apply time period factors, by trip purpose, to generate the time period matrices for subsequent use in the assignment process. The time period factors indicate the proportion of the daily travel, by purpose, undertaken during the time period to be modelled and are usually derived from travel surveys. For strategic network assignment modelling, the peak period trip matrices are generally for either a one-hour or two-hour time period, depending on the requirements of the analysis.

Where the travel forecasts are to be used to inform the public and/or the private sector regarding an investment in transport infrastructure in areas where there is evidence of congestion it is essential that multiple time periods of the day be modelled, at least three, AM peak, PM peak and off-peak.

When time period factors are used to determine time period demand, the challenge here is to consider how these factors change over time where congestion effects cause peak spreading. This is discussed further in section 3.2.5.

3.2 The four step transport modelling process

A commonly used model structure is the ‘four-step’ transport modelling process. The steps in this process are shown in Figure 5 and described below. An important feature of the four-step modelling process is the iterative feedback of costs arising from trip assignment to trip distribution and mode split. By iterating between the last three steps (trip distribution, mode split and trip assignment), it is possible to replicate the impacts of congestion on travel costs. This iterative process ensures a balance between the final trip pattern and the costs by which it is derived.

Figure 5: The four-step transport modelling process The four-step transport modelling process

The four step model is a framework that represents the scope of functions that are typically represented in transport models. The emphasis given and complexity of individual steps, and interactions between the steps is part of the model design task. In contexts where there are substantial changes in accessibility envisaged, or the modelling has been focused on a particular corridor, it may also be important also to model changes in trip frequency (whether trips as defined within the model scope are made or not) and possibly changes in land use and thus expanding and more fully integrating the trip generation step. In other cases there may be a need to represent changes in when trips are made, in which case a fifth ‘step’ may need to be incorporated to this standard framework.

3.2.1 Step 1 – Trip generation

Trip generation is the procedure whereby land use, population and economic forecasts are used to estimate how many person trips are produced within, and attracted to, each zone. Trip generation uses average trip rates for the study area to estimate the quantum of trips undertaken by various trip purposes such as:

  • Home-based work trips (such as work trips that begin or end at home)
  • Home-based shopping trips
  • Home-based education trips (such as from home to primary, secondary and tertiary education)
  • Home-based recreation trips
  • Home-based other trips
  • Non home-based trips (trips that neither begin nor end at home)
  • Other non home-based trips (such as service trips and business trips).

The productions and attractions developed in the trip generation step are usually termed ‘trip ends’. Generally for home based purposes, trip productions depend on the population characteristics of a zone, while trip attractions relate to employment parameters, school places, retail floorspace healthcare provision, etc. In some projects it is important to consider alternative land use forecasts, such as to assess the implications of alternative urban form proposals for the transport networks, or to consider the robustness of the performance of particular schemes to uncertainties in future land patterns.

Table 4 provides examples of the type of data that could be used in this step.

Table 4: Demographic and land use variables used in urban transport modelling
Demographic data Detailed demographic data by transport zone for the study area including:
  • projected resident population by age cohort, and
  • current age cohort of population.
Land use data Data on land use would include:
  • quantities of land / size of developments required for various uses (e.g.retail, health, leisure, residential, commercial, etc) to meet projections of population and employment
  • analysis of land use by industry type
  • proportion of commercial land occupied,
  • register of available open space within study area, and
  • current and projected school enrolments by level: kindergarten, primary, secondary and tertiary at transport zones.
Economic activity data The following data may be used to provide an economic activity base for the forecasting future trip-making levels:
  • current employment levels at transport zones
  • forecast employment by industry and/or occupation classification
  • anticipated employment growth within transport zones
  • assumptions about the employment generating capacity of the region, and
  • estimated input-output function of the study area.

The reader should also be aware that trip generation models might also incorporate variables related to the transport network being modelled – both roadway and transit. This will build into the model the capability of representing induced travel demand in the form of more or less travel being made and by changes to land use patterns.

Accessibility is the main influence of transport provision on land use. Locations with better transport links tend to provide companies with better access to their markets and a larger pool of employees. An individual’s choice of where to live is, in part, influenced by their ease of accessing work places, schools and other amenities. Land use transport models represent the rate of development and location choice decisions as a function of accessibility. In some cases other indicators, such as environmental conditions (derived from traffic flows and vehicle emission rates) are also represented (e.g. see Whitehead et al). From the perspective of the transport model, it is important to establish accessibility, preferably representing all transport modes and weighting land use as a variable in the trip generation model (principles are discussed by Hansen (1959)). While the practice of incorporating accessibility to trip generation model is not common in Australia, it should be considered when the issue of induced demand becomes important for the model scope (see Section 3.4 for further discussion on induced demand).

Zoning system

A key issue to be resolved when developing transport models is the transport zoning system. Transport zones should ideally contain homogeneous land use (for example, solely residential, industrial or commercial use or parking lots) and they should not cross significant barriers to travel (such as rivers, freeways and rail lines), and should have reasonably homogeneous access to the modelled transport systems. In this context, transport zones should match, as far as practically possible, Australian Bureau of Statistics (ABS) Statistical Area boundaries as defined by the Australian Statistical Geography Standard (ASGS). In general practice, transport zones will be aggregations of the Statistical Area Level 1 (SA1) or mesh block boundaries.

Land uses with specific trip generation characteristics, which cannot be adequately described by the trip generation equations derived for other land uses, should be coded as separate zones (such as airports, ports, universities, hospitals, intermodal terminals and shopping centres).

In theory, the accuracy of a transport model should increase with the number of transport zones. However, it may be difficult to obtain reliable input data (employment, population) at a highly disaggregated level. The trade-off, therefore, is between the accuracy of the transport model and the practicality of having existing input data for transport model development at the chosen level of geography, and also being able to forecast the input data.

Another point to be considered when defining the zoning system is the highway and public transport network detail required to support the defined zoning system. A highly disaggregated zoning system will require a concomitantly disaggregated network to ensure all trips from all transport zones can access the transport network (via the centroid connectors) and that there is a reasonable concordance between the modelled and observed traffic volumes.

Transport zone centroids are defined as the ‘centre of gravity/activity’ of a transport zone and centroid connectors are used to load the trips from a transport zone onto the modelled transport network. Centroid connectors should represent, as closely as possible, zonal ingress and egress at reasonable access points to the network. Ideally, centroid connectors should not be connected to intersections: it is preferable to connect them to mid-block points in the modelled transport network.

3.2.2 Step 2 – Trip distribution

Trip distribution determines where the trip ends - developed in trip generation (Step 1) - will go. These trip ends are linked to form an origin–destination pattern of trips through the process of trip distribution. The logic behind trip distribution is that a person is more likely to travel to a nearby transport zone with a high level of activity (such as employment, shopping or recreation) than to a more distant zone with a low level of activity.

There are several approaches for trip distribution such as growth factor, gravity model, entropy-maximising approach, intervening opportunities (Ortúzar and Willumsen, 2011). The most commonly used procedure for trip distribution is the gravity model [1]. The gravity model takes the trips produced at one particular zone and distributes them to other zones based on the size of the other zones (as measured by their activity or trip attractions) and on the basis of some impedance to travel between zones. Thus, the number of trips between zones is usually related to the degree of land use and activity within each zone and the ease of travel between them.

Recent developments in trip distribution have seen the implementation of logit based destination choice models. While destination choice models may use different mathematical functions to gravity models, they share many of the same characteristics and similar to the gravity model, can be either singly or doubly constrained.

Impedance can be measured in several ways. The simplest way is to use either actual travel distance (km) or travel times (minutes) between zones as the measurement of ‘impedance’. Alternatively, by ascribing a value of time and a vehicle operating cost rate to travel time and travel distance respectively, together with any tolls paid, a ‘generalised cost of travel’ [2] can be used as the ‘impedance’. It is necessary that the gravity models or destination choice models should be applied to the network costs representing the effects on future travel patterns. An approach assuming that the base year travel patterns do not change in the future is unlikely to be realistic.

It is usual to have a separate gravity model developed for each trip purpose, since different trip purposes exhibit different trip distribution characteristics. The outcome of the trip distribution step is a matrix of trips from each transport zone to all other transport zones.

3.2.3 Step 3 – Mode choice

Mode choice allocates the origin–destination trips derived from trip distribution (Step 2) to the available travel modes, by trip purpose. This step estimates the choice between travel modes based on the characteristics of the trip maker (income, car ownership, age), the trip itself (trip purpose, the origin and destination) and the characteristics of the travel mode (fares, vehicle operating costs, travel time, parking availability and cost, reliability). The outcome of this step is an estimate of travel by all available travel modes between all transport zones, by the separate trip purposes.

The development of mode choice models usually relies on information such as the observed mode choice (from survey data or other sources), the characteristics of people undertaking the travel (age, employment status, currently studying and at what level, if they hold a licence) and the characteristics of the travel modes (availability, frequency, price, reliability).

Travel modes for personal travel include:

  • Walking
  • Bicycle
  • Car
  • Bus
  • Tram
  • Train.

Mode choice can be performed before trip distribution (trip-end mode choice model) or after trip distribution (trip-interchange mode choice model). Alternatively, trip distribution and mode choice may be performed simultaneously using a composite cost function (Otuzar & Willumsen 1994).

Trip-end mode choice models split the total demand for travel for each transport zone by the available travel modes. The mode choice in this case is based on the attributes of the trip origin (that is, ease of access to each mode and the ability or inclination to use a particular mode). The trip-interchange mode choice models split the origin–destination travel (including intra-zonal travel) between the available travel modes by responding to the specific service characteristics of the available travel modes. In this approach, the number of trips by travel mode is estimated on the basis of the relative utility (or disutility) of travel by different modes, as perceived by the trip maker.

The most commonly used form for mode choice is the ‘logit’ model – a discrete choice model - which is based on the assumption that an individual associates a level of utility (or disutility) with each travel mode in undertaking travel between transport zones. In practice, a generalised form of logit models known as nested logit models are used to model mode choice. Figure 6 below shows an example structure of a nested logit model.

Figure 6: Structure of a nested logit mode choice model Structure of a nested logit mode choice model

The example above incorporates three broad modes of travel: car, public transport and non-motorised transport. The car mode is split into car driver and car passenger. Public transport is split into three separate elemental modes, depending on the mode of access taken to use the transit services (note that this model does not distinguish between rail or bus services). Non‑motorised transport is split into bicycle and pedestrian modes. The application of nested logit models is documented in a readable way in Hensher et al (2005).

It should be noted that traditional logit (MNL and nested) does not allow for differences in user specific taste. Modelling user heterogeneity and their difference choice patterns requires the development of specific models for each of the demand segments or the use of mixed-logit models, which are not the standard practice.

A brief account of discrete choice models, in general, can be found at Appendix B. For further information on the development, estimation and application of discrete choice models, the determination of elasticity values from discrete choice models and the application of the models in project evaluation, see: Ortuzar and Willumsen (2011); Taplin, Hensher and Smith (1999); Hensher and Button (2000); and Louviere, Hensher and Swait (2000).

3.2.4 Step 4 – Trip assignment

Trip assignment assigns the various mode-specific trip matrices, by trip purpose, to the alternative routes or paths available across the transport network. Public transport trips are assigned to the public transport network (where path choice includes all public transport modes); and vehicle trips and are assigned to the highway network. This step provides an indication of the likely distribution of travel across the available transport network.

Trip assignment results can be used to:

  • Identify and assess deficiencies in a transport network
  • Assess the transport network performance
  • Evaluate the impacts of transport infrastructure proposals
  • Evaluate alternative transport system and land use policies
  • Provide inputs to economic appraisal.

Section 3.6 Network models provides further details on trip assignment.

3.2.5 The five step transport modelling process

At its simplest, the four step model involves the assignment of daily highway demand into the network. For daily traffic assignment, practitioners would usually estimate the daily capacity for a link by applying a factor to its hourly capacity, based on the relationship between daily traffic and hourly traffic counts. This method is simple to develop and operate and may be suitable for small regional towns with low traffic congestion. The approach does not distinguish marked differences in traffic pattern between AM and PM periods. In larger towns where there is congestion, time period factors are applied to disaggregate demand between individual modelled periods and thus representing variations in trip patterns, such as typical commuting patterns towards employment centres in the morning. This provides a significant improvement over the all day assignment, since it is able to represent traffic demand and congestion over different time periods. The time period factors are defined by travel purpose and distinguish direction of travel to and from home and are typically derived from household surveys.

A five step transport model would incorporate an additional step called time period model, to split the daily demand between time periods. Currently, almost all Australian models have fixed time period factors which are applied to split daily demand into AM, Inter-peak, PM and Off-peak periods. However, the assumption of fixed factors of time period for future year models might overestimate the level of traffic congestion during the peak periods as travellers would consider changing their time of departure to avoid traffic congestion. 

Recent developments in five step transport models incorporate the capability to model the departure time choice of a traveller in response to changes in travel conditions. In many circumstances, travellers will be more likely to change their time of travel, rather than their trip destination or the mode of travel, in response to changes in travel conditions. The incorporation of a departure choice step, for example, can better enable the modelling of the peak spreading phenomenon. As discussed above, the use of fixed trip timing assumptions can result in overstatement of modal and distribution responses in forecasting where there is significant peak period congestion.

Figure 7 An example of a five step model including travel time choice An example of a five step model including travel time choice

When fixed assumptions are made about the proportion of trips at different times of day for different purposes, consideration should be also given to other factors such as trip length, reflecting for example the earlier departure time frequently observed for commuters making long journeys. Care is required also to reflect the objectives and scope of the model; the time a long journey traverses the main modelled area may be of more interest than the departure time.

It is useful to distinguish behaviours about scheduling activities, which may be associated with changing travel times by one or more hours, from those about how best to make a journey, which may involve adjusting travel times by a few minutes.

Approaches to represent the former, macro-time period choice, are mathematically similar to those used for mode choice modelling although the linkage of individual trips into tours starting and ending at a travellers home is usually advisable to reflect the interactions between different trips. Hess at el 2008 discusses methods and Fox et al 2014 provides a commentary on different approaches that have been applied.

Experience on the latter, micro-time period choice, is more limited. The most common approach is to apply a choice model for departure time that incorporates a scheduling penalty for arrival or departure before or after the ideal time. However integration of this in an area wide model also requires network models that can reliably differentiate and represent the differences in network conditions and travel times for small differences in departure time. The applications have therefore tended to be localised and context specific.

3.3 Demand elasticities

Analysis using demand elasticities is based on the assumption of a direct relationship between the change in a policy-dependent variable and the corresponding change of a particular transport choice. The elasticity of demand, with respect to a given parameter (explanatory variable) may be seen as the percentage change in quantity demanded resulting from a one per cent change in the value of the parameter.[3] Both the magnitude of the percentage change and the sense of that change (positive or negative) are important. Direct elasticity refers to the change of demand for one transport service (or mode) in terms of the change in a parameter affecting that service (or mode). Cross elasticity refers to the change of demand for a service or mode resulting from the change in a parameter affecting a different (competing) mode or service. Thus, for example, the change in patronage on a rail service as a result of the increase in the fare charged for that service could be estimated using the direct demand elasticity for rail with respect to fare. The increase to an alternative mode (such as bus) resulting from passengers switching modes could be estimated using the cross elasticity of demand for bus with respect to rail fare.

Similarly, the elasticity of demand with respect to income measures the proportionate change in demand resulting from a proportionate change in the income of the consumer. As consumers have more income, their choices may expand or change.

Elasticity models are used to estimate these effects. If the value of the parameter is P and the demand (such as number of trips) is Q, then the elasticity ƞ is given by:

η = Δ Q Q Δ P P = P Q δ Q δ P

The partial derivative δ Q δ P is used to indicate that all the other explanatory variables (parameters) that can affect demand are assumed to be held constant. This derivative is the slope of the demand curve that relates to Q and P. The slope of the curve may vary along its length so the value of ƞ as defined by the equation above is only valid for the particular point at which the slope is measured. It should strictly be called the ‘point’ elasticity.

Elasticity values can be derived for many parameters, such as fares, costs and charges, travel times, service reliability, service frequency or service quality (e.g. passenger comfort or possibility of damage to or loss of commodities). Provided the relative change in the value of each of these parameters is small, the overall change in demand can be estimated as a simple sum of the individual change estimates

Δ Q Q = η A Δ P A P A + η Β Δ P Β P Β + η C Δ P C P C +

for parameters PA, PB, PC, etc. Models of this type have been used for many years by transport operators. They provide a quick estimation of the likely effects of proposed service changes, based on previous experience, as long as the service changes are incremental.

Point elasticity as defined by equation above is rarely available for practical applications, because knowledge of the mathematical shape of the demand curve would be required to determine it – and this is seldom the case. Rather, most elasticity values used in practice are arc elasticities, estimated by considering measured differences in demand at different values of a given parameter. If the demand Q1 corresponds to value P1 and demand Q2 corresponds to value P2 of the parameter, then the arc elasticity ƞ is given by

η - = ( Q 2 - Q 1 ) ½ ( Q 2 + Q 1 ) ( P 2 - P 1 ) ½ ( P 2 + P 1 ) = Δ Q Δ P P - Q -

where P and Q are the respective mean values. Given the non linear shape of logit and most demand curves, the arc elasticity is commonly used in log form as below, to be applicable across a slightly wider range of cost changes:

η = L o g Q 2 Q 1 L o g P 2 P 1

The demand to use a particular mode or service can be affected by changes in travel parameters applying to competing modes as well as changes in the parameters of the given mode. The effect on demand for use of mode H of changes in a parameter PG on mode G is given by the cross elasticity:

η H G = Δ Q H Q H Δ P G P G

Elasticity values are generally estimated from time series data. For instance, transport operators may collect information on patronage or commodity movements over time and this can be used to suggest the effects on demand stemming from changes in transport costs or other service parameters (for both their own operations and those of their competitors).

Elasticity values for passenger transport by transport mode may be found in reported studies such as Oum, Waters and Yong (1992) and Luk and Hepburn (1993). A substantial database of elasticity values has been compiled by the Bureau of Transport and Regional Economics and is available on the BITRE website (www.bitre.gov.au) under databases and products. Taplin, Hensher and Smith (1999) provided a discussion on the conceptual and theoretical requirements on the estimation of elasticity values. When applying the elasticity values from literature to any project, one should examine the applicability of these values considering the place, time and the source of data used when they were derived, and their valid range.

Certain conventions need to be followed when interpreting published values of elasticities. In the case where ƞ denotes a demand elasticity with respect to parameters such as cost or trip time, the value of the elasticity is negative – an increase in the price of a service would normally be expected to lead to a decrease in the demand for it, but common practice among economists is to quote only the magnitude (modulus) of the value (as a positive number, with the implicit assumption that the value is actually a negative number). Cross elasticity values are generally positive – an increase in the charges for use of a competing service would be expected to result in an increased use of the other service. The dimension of demand should be checked for example whether it is expressed in number of trips or trip kilometres before applying the appropriate elasticity.

3.4 Induced Travel Demand

The issue of induced travel demand has been discussed at length in literature (U.K Department of Transport (1993), Abelson and Hensher (2001), Litman (2008) and Ian Wallis Associates for Department of Transport Victoria (2009)). The Victorian Auditor-General’s Report (Victorian Auditor-General’s Office, 2011) recommended road authorities to assess the significance of induced traffic for all major road projects and consider it when forecasting traffic and estimating the economic benefits. Subsequently Victoria Department of Transport (DOT) (2011) released a draft position paper on induced travel demand.

3.4.1 Sources of induced demand

Induced travel demand refers to the impacts of new transport projects and services in encouraging some people to switch routes, modes or time of travel to take advantage of the improved travel times and service levels. In addition, induced demand can refer to the tendency of some people to travel more when travel conditions are improved. Longer term effects may include some households and businesses locating close to the new or improved transport infrastructure and services, and/or locating further away from their destinations. Induced travel demand can arise from both road and public transport (PT) projects (DOT, 2011). Overall, induced demand resulting from a transport improvement may include six following components:

  1. Change in route: The transport improvement may make one route faster. Traffic travelling between A and B switches to the improved route, resulting in induced trips on the improved route although not necessarily in the network as a whole.
  2. Time of day: The transport improvement may improve travel speed at one time of day relative to another (e.g. peak hour vs off-peak). Trips travelling between A and B switch from the slower to the faster time of day (e.g. from off-peak to peak), resulting in induced trips at the faster time of day (e.g. peak hours) although not necessarily more trips on the network as a whole over the day.
  3. Changes in mode choice: An improvement in one mode will cause some people to switch to it from other modes, For example, a road improvement would cause public transport users to switch to car travel, resulting in induced car trips. This is the most commonly achievable variable matrix approach in most capital city transport models (SKM 2009);
  4. Trip redistribution: Trip destination change. Improved travel speeds may encourage people to switch from a close destination to a more distant one that becomes more attractive. This results in induced travel kilometers of travel.
  5. Trip generation:  As the transport improvement makes travel less costly, new trips may be made that were not made previously on any transport mode.
  6. Land use changes: In the longer term, the new or improved part of the transport system may encourage higher population and business activity near the improved facility, and/or encourage households and firms to locate further away from their destinations - both contributing to increased traffic flows (see section 3.5 for discussion of modeling of land use and transport interaction).

Depending on the scale, large projects such as a new freeway may generate all six components of induced demand, while small projects like a small town bypass may cause only a change of route. The extent of induced demand was further discussed by DOT (2011) as follows.

Evidence of the induced demand effects of trip generation and trip redistribution for urban transport projects is limited. In terms of impact on travel time, the extent of new trip generation and trip redistribution in peak periods appears to be negligible. The impact on off-peak travel times is considered to be smaller still.

In large cities where rail’s share of the commuting trips to the central area is substantial, it was found that modal shift from public transport can account for up to half of the estimated induced traffic on a road corridor. As a proportion of total screenline traffic, literature suggests that induced demand as a result of modal shift can account for between 2% to 3% of total screenline traffic flow.

Studies examining different road schemes generally find trip re-assignment to be a substantial factor contributing to induced demand, which could range from 10% to 25% of total screenline traffic depending on location-specific factors such as time savings created by the specific project and the extent of congestion on existing/alternative routes.

Prior to the construction or enhancement of a major road link (or public transport service), some travellers may choose to travel outside peak periods to avoid congestion; however, they may reschedule their trips if the time advantage of doing this is no longer as great. This means a road (or public transport) improvement could induce travellers to retime their trips, adding to the numbers travelling during peak periods. Surveys have found that between 10 per cent and 30 per cent of respondents reacted to road improvements by reverting to travel on those roads during peak periods. Trip retiming could be a significant source of induced demand during peak periods. However, trip retiming does not add to the increase of a total daily trips or VKT.

New roads and upgrades to existing roads can alter land use patterns, and the location decisions of both firms and households. In particular, activities that are reliant on accessible locations would tend to increase around new road developments such as distribution/warehousing activities, large mall and superstore developments, and offices requiring good access for employees and visitors. It is likely that significant public transport improvements would also encourage land use changes over the medium to long term. However, the empirical evidence of induced traffic from land use changes is scarce. Nevertheless, induced demand from land use changes would be largely a medium to long term phenomenon.

3.4.2 When is Induced Demand Significant?

Induced travel demand effects are of greatest importance for the demand modeling economic appraisal of transport projects in networks with:

  • a high degree of congestion (typically in urban areas, especially at peak periods); and/or
  • high elasticity of demand (typically in urban areas, especially where alternative modes offer strong competition); and/or
  • relatively large changes in travel costs (typically for larger schemes providing substantially enhanced capacity).

For public transport network improvements, induced demand effects are also most significant when similar conditions apply – that is, when demand is elastic and increases in response to improved service, and when the service is already congested or crowded.

3.4.3 Modelling Induced Travel Demand

The Transport Analysis Guidance (TAG) (Department for Transport London, 2014) suggests three broad approaches to representing travelers’ response to cost:

  • a fixed demand approach, or Fixed Trip Matrix (FTM) approach, in which demand is independent of cost, and the trip matrix is adjusted using trip ends and no behavioural model is required;
  • an own cost elasticity approach where demand in each cell of the trip matrix can vary, but the source of any variation is limited to the corresponding cell of the cost matrix only; or
  • a full variable demand approach, or a Variable Trip Matrix (VTM) approach, where demand in each cell of the matrix can vary according to demand in other cells of the trip matrix and costs in all cells of the cost matrix. This is usually implemented using discrete choice models.

Use of the FTM approach is only valid for generally small or short term schemes where it can be demonstrated that changes in travel cost will not generate a noticeable change in demand. This method should only be used if the network improvement would only generate route changes. As such, FTM models are inadequate for most transport schemes which are aimed at resolving congestion or relieving overcrowding on public transport. The FTM modelling could also materially overstate the economic benefits a scheme could deliver (see Abelson P. & Hensher D., 2001 and Litman, 2008).

Own-cost elasticity models do not constrain total demand according to the size of the population. This means they are not adequate for representing either the transport market as a whole, or modes with a high share of overall travel, such as car. However, they have advantages over choice models when analyzing rail schemes as it is much less expensive to build. DOT (2011) provided several elasticities with respect to travel time and network capacities. Own-cost elasticity” models are in common use in other countries. They can produce accurate forecasts if based on detailed observed passenger origin-destination data such as from smart card or other detailed ticket data.

The VTM approach is required to assess induced demand. Generally a four step transport model would provide variable demand and be able to address at least the issue of induced demand with respect to redistribution of trip (by trip distribution module), change of mode (by mode choice module) and change of route choice (by trip assignment module). The level of accuracy would depend on model design.

Where strategic transport models for Australian major cities do not model time of day choice (Planning and Transport Research Centre, 2014), they are unable to assess the induced demand due to travelers who change their time of travel. There are two distinctly different aspects of time of day choice:

  • Macro time period choice - representing the choice between broad time periods within a day, e.g. whether travel is in a peak period or an off-peak period
  • Micro time period choice (peak spreading) - representing departure time choices of a few minutes.

Variable demand models only usually include macro time period choice to represent transfer of traffic between broad time periods. Logit type choice model can be used for both macro and micro time period choices. For example, macro time period choice (the allocation of trips between broad time periods – see Section 3.2.5) takes the form:

T i j m s = T i j m * e x p ( - λ t i m e G i j m s ) t e x p ( - λ t i m e G i j m s )

Where

T i j m s is the number of trips between zones i and j by mode m in time period s.

G i j m s is the disutility or generalised cost of travel between zones i and j by mode m in time period t, which may typically be peak and inter-peak and λtime is the choice sensitivity parameter for the time period step.

Currently, most Australian transport models were developed with static trip generation rates, i.e. trip generation rates do not change with improvement of transport system. The development of a dynamic trip generation module with trip generation rates being a function of transport accessibility (Ortúzar, J. and Willumsen, L., 2011) may enhance models in this aspect. The cost of development of this component needs to be balanced against the benefits of having it. Separately DOT (2011) indicates that there is a general agreement among transport planners that entirely new trips represent a relatively small share of the increased traffic appearing on a new or improved highway facility.

Where the variable demand (VTM) model is used to forecast responses to scheme related cost changes, there is a need to ensure satisfactory convergence of the demand-supply iteration in addition to the network model convergence (Section 5.9) particularly when estimating user benefits. Where the model is used to estimate user benefits (see section 3.4.4 below), the demand/supply gap should be measured as follows:

a C a n D a n - D a n - 1 a C a n D a n

Where

C a n is the generalised cost for matrix cell a (spanning origin, destination, mode, time and segment) for iteration n

D a n is the demand forecast by the demand model using cost C

A demand supply gap of less than 0.001 may be satisfactory in most cases, but where the scheme is small relative to the model tighter convergence may be necessary. Assessed user benefits from previous demand/supply iterations of the model can indicate the stability of the user benefits calculation.

The transport system and land use system are closely interrelated. An improvement in the transport network would improve accessibility which would drive change in land use redistribution that in turn would redistribute trips and sometimes generate additional trips. The induced demand due to land use changes requires using a land use transport interaction model – see section 3.5.

The following table summarises the recommended modelling approach – FTM or VTM - to undertake to model the different types of induced demand.

Table 5: Recommended modeling approaches to model induced demand
Type of Induced Travel DemandModelling Approach
Changes in route FTM
Changes in time of day VTM
Change in mode choice VTM
Change in trip redistribution VTM
Change in trip generation VTM
Change in land use VTM

In the next version of the ATAP Guidelines, detailed guidance will be provided on the recommended application of the FTM and VTM approaches.

3.4.4 Measuring User Benefits

Part T2 Cost-Benefit Analysis of the guidelines, discusses the measurement of user benefits. Section 7.4 therein indicates that travel demand models are required to estimate user benefits in complex urban settings, and provides user benefit equations to be used. It indicates that the user benefit calculations should be undertaken within the travel demand model as follows:

  • The user benefit calculations first need to be undertaken at a disaggregated level for each origin-destination pair, for each mode, for each time period and for each forecast year
  • The disaggregated results are then aggregated to yield overall use benefits.
    • aggregating across the entire demand matrix, i.e. across all origin-destination pairs
    • repeating the process for all modes and time periods
    • repeating the process for each model forecast year.

Where induced demand is expected as a result of a transport improvement, the Variable Trip Matrix (VTM) methodology should be applied in demand modelling and user benefit estimation. A VTM approach should be used, in conjunction with a variable demand model as discussed above. The VTM approach differs from the Fixed Trip Matrix (FTM) approach in that, for a given forecast year, the demand in the Option Case is usually higher than that in the Base (Do Minimum) Case.

Consistency must be maintained in calculating user benefits across all origin destination pairs, modes and time periods represented in the transport model, as discussed further by Jones (1977). Issues that may need specific consideration include the following:

  • Introduction of new modes, or where the changes in cost are large and the demand curve cannot adequately be approximated by a straight line. In this case approaches include the use of composite cost derived for more aggregate representation of choices, or through testing a sequence of ‘interim’ supply curves providing a better approximation to the demand curve.
  • Changes in land use, whether forecast by an integrated land use model, or with development constrained without the intervention. In this case the land use related costs are not included in the cost per trip and the user benefit calculation must be constrained to fixed land use assumptions, with separate calculations undertaken on the benefits of the development (or land use changes).
  • Instability or lack of convergence in the transport model (see also section 3.4.3 above). Testing with for example one additional transport model iteration can illustrate the level of uncertainty in the user benefit calculation. In some cases where model noise is attributable to issues remote from the intervention, actions to constrain the changes in cost represented in the model (either by excluding remote origin-destination movements from the user benefit calculation or constraining the travel cost changes represented in the transport networks may be justifiable.

One of the reasons for applying the ‘rule of a half’ calculation is the ability to report separately what changes the user benefits derive from (e.g.travel time, monetary cost, perceptions of travel quality). Similarly the value of travel time savings is judged to be greater for business travellers than for other travellers and these should be distinguished in reporting. It may be that particular population segments need to be distinguished, depending on the objectives the intervention is intended to address.

3.5 Land use transport interaction (LUTI)

A key input to strategic transport models is land use information. There is little argument about the fact that there is a relationship between transport demand and urban structure. The opposite relationship, that urban development is related to the transport supply, is also accepted yet not fully understood. It is generally accepted that transport accessibility is an important ingredient for understanding the potential for land use development. However, there is still some uncertainty as to the most appropriate method of implementing transport land use interaction models given the reoccurring tension between theoretical preferred options and practical implementation. The aim of this section is to provide reference material on the evolving area of land use transport interaction. More guidance will follow in the next update.

3.5.1 Overview of types of land use transport interaction models

The following section provides a brief overview of the different approaches used for land use transport interaction models, as summarized below.

Structure \ ModeStaticDynamic
Linked Separate Land use and transport model, run iteratively to convergence representing single forecast year Separate land use and transport models, run sequentially, say in 1 or 5 year increments, representing dynamic evolution over time
Integrated Interaction of land use / activity patterns directly mapped to transport needs, run iteratively or through simulation if disaggregate, to equilibrium E.g. using system dynamics software, directly linking land use / activities to transport, and representing lags in rate of response

Static equilibrium models

Static equilibrium LUTI models are analogous to the four step transport model in that they are generally macroscopic in nature. Early LUTI models such as the Lowry model (Lowry, 1964) used gravity formulations or input-output formulations. Static LUTI models are usually directly linked to four step models and can be used to estimate equilibrium patterns of land use by iterating with a four step models, which in turn create measures of accessibility to be input to the land use model. In a four step model, land use changes are forecast first based on planned developments, with no direct input from the transport model. In static equilibrium LUTI models, the land use model includes functions representing how accessibility affects land use, and the land use and transport models are run iteratively to an equilibrium with land use forecasts input to the transport model and accessibility (travel cost) forecasts input to the land use model. Static equilibrium models can be implemented by combining separate land use and transport models through an iterative process, or can be fully integrated.

Dynamic models

Dynamic models (Wegener, 2013) are generally classified as either macroscopic (General spatial equilibrium models) or microscopic (agent based models). They are based on random utility theory and theories of competitive markets. In some instances these are referred to as based on systems dynamic theory, that is representing the time dependent interactions and lags between land use and transport related systems. They can be fully integrated with transport models rather than being exogenous inputs to each other. Dynamic agent based models are more analogous to activity based transport models.

Linked versus integrated models

In fully integrated models the destination choice is undertaken within the land use part of the model. Therefore, the land use and transport components of the LUTI model cannot be separated. In contrast, in linked LUTI models the destination choice is undertaken in the transport part of the model, as in a regular four step model. This allows the land use and transport models to be separate from one another but link to each other in an iterative manner (DfT, 2014).

The advantage of fully integrated models is that the travel formulation and land use formulation is entirely internally consistent as the processes are endogenous. However, having these processes fully integrated assumes a high level of confidence in the land use model. Fully integrated models may find it difficult to reach unique solutions, which then pose a challenge for undertaking economic assessments of infrastructure projects.

Linked models may lack the internal consistency of fully integrated models; however, they are more flexible in their application in that the interaction between the transport and land use models can be turned on or off.

3.5.2 Use of LUTI models in Australia

The Planning and Transport Research Centre (2014) provides several references to the application of land use models in part of Australian cities. These include:

  • Metropolitan Land Use Forecasting System (MLUFS) by the Ministry for Planning, Western Australia (1996)
  • UrbanSim application in Logan City Queensland (Brits, 2013)
  • Large Scale Urban Model (LSUM) applied in South East Queensland to simulate potential future patterns of population and jobs at a spatially disaggregated level (Stimson et al., 2012)
  • Queensland Small Area Model (QSAM) constructed by Demographics Australia for use by the Queensland Government to project total population and dwellings for up to 500 small areas within an urban Local Government Area (Wilson, 2011).

Apart from UrbanSim, a dynamic disequilibrium model with capability to connect with a four step model, the other land use models are static equilibrium, without location choice or capability to link to transport model.

In most transport jurisdictions within Australia transport and land use models operate independently from each other and only an occasional interaction usually when assessing major infrastructure projects. When applied, the LUTI models generally have the land use model and transport model linked rather than integrated. There are differences in view with some practitioners adopting a take the form of static equilibrium models with the land use model and transport model linked, but not necessarily run to a convergence. There is not complete agreement whether it is sensible to model a converged land use transport interaction model, or to represent the evolution of land use over time using a ‘systems dynamic’ approach.

Given that land use and transport models have been separately developed, linkage of these tools is relatively low cost route to establish a LUTI model. The key actions are as follows.

  • Establish a file management structure to control the iterations between transport and land use models. For a static model this will include a process to measure and test convergence, for which it would be adequate to measure stability of transport model metrics. For a dynamic model, the process would be arranged to operate the land use model typically for each year through the modelled period. The transport model may be run in say 5 year intervals, taking inputs from the relevant forecast year of the land use model and providing outputs to the land use model for the subsequent few years. In both cases the transport model would be run to convergence.
  • Link land use model forecasts of population and employment to provide inputs in a suitable form for the transport trip generation model.
  • Link the transport model generalised cost outputs in a suitable form to represent accessibility changes input to the land use model.
  • In some instances there can be other interfaces, such as environmental indicators based on traffic volumes in an area, depending on the capabilities and relationships represented in the land use and transport models.

The additional iterations between transport and land use model require consideration in verifying and interpreting forecasts. In a dynamic model for example the rate of change over time may need to be considered, in addition to the final outcomes. It is also often helpful to run the models separately to separate the effect of the individual transport and land use relationships represented in the LUTI model and this better explain the forecast outcomes.

3.5.3 Appropriate use of LUTI models

Static equilibrium LUTI models provide a tried and tested method of modelling the interaction between transport and land use, albeit with some significant behavioral shortcomings. These static models will likely be the LUTI model of choice when combining with a four step models for the purpose of producing land use forecasts to be used in the appraisal of major transport infrastructure projects.

Relative to a dynamic model, the static model is simpler to implement. It also provides outputs that can be interpreted in appraising the impacts of transport schemes more easily to allow the analyst to explain the forecast impacts of the transport intervention on land use and the wider economy.

However, for the purpose of understanding transport and land use interactions for the purpose of developing policies a ‘system’ dynamics’ approach may provide more insight to policy makers. Land use changes (and some transport behavior) evolve over years and decades. A dynamic model more realistically represents the evolution of changes. The added complexity for the analyst is to investigate a number of forecast years to understand these interactions. This is complicated by the absence of equilibrium principles in the model process.

A particular value of LUTI models is the capability to represent internally consistent visions of the future, or different scenarios. The use of different scenarios helps the decision maker understand the factors that cause pressure for development or stress in the transport systems and how these relate. They also provide a basis to help interpret when particular interventions may be appropriate, and what drivers may cause the need for intervention to come forward or recede in time.

3.5.4 Validation and data requirements

For static land use models, regression analysis looking at the correlation between planning data (residential intensity, employment intensity) and accessibility (as derived from the transport model) draw upon the information that would be required for conventional four step transport modelling. In some cases the land use model tools have been developed drawing on a range of research evidence to establish suitable functional relationships with outputs available from the transport model. Usually additional segmentation is adopted distinguishing employment types than are commonly applied to transport models.

More sophisticated land use models consider also constraints of the planning system, the extent of land available for development and in dynamic models the speed of change that can be achieved. Particular data requirements in these cases are the extent of land that may be developed, and for what purposes, over the forecasting period.

Land use changes arise from a range of sources, beyond those related directly to the transport systems. It is important therefore to take a dynamic approach to testing and verifying the performance of the land use models. This would first compare the rate and nature of forecast population and employment growth over time, comparing this with historic trends and existing policy based plans. Secondly the responsiveness of the land use changes to transport changes should be considered, reflecting both the scale and location of forecast changes.

3.6 Network models

Transport network models generally comprise a combination of interconnected links (sections of roads) and nodes (representation of intersections) and are termed ‘link-based’ network models. Public transport services are represented in part by the sequence of nodes (stops) and links traversed. The transport network model, or ‘supply side’ component of a transport model is intended to reflect, as accurately as possible, the actual available transport network, by incorporating the following link attributes:

  • Distance
  • Free speed on the link or travel time
  • Capacity, as related to the number of lanes or available train paths
  • Direction indicator (one-way or two-way link)
  • Volume delay function.

The transport network model should be sufficiently refined to support the adopted model zoning system and, as a minimum, should cover the following road classifications:

  • Freeways and tollways (coded as one-way links with associated ramps coded separately)
  • Divided and undivided arterials
  • Collector roads
  • Local roads (the extent to which these are included is at the discretion of the modeller and the requirements of the particular study)
  • Rail lines (freight and public transport)
  • Ferry lines and services
  • Bus-only corridors.

3.6.1 Highway traffic assignment

One of the most commonly used assignment techniques is the capacity restrained assignment that involves the allocation of traffic to routes according to the ability (or capacity) of the route to accommodate traffic. It should be noted that the capacity of a route is not like the capacity of a bottle; rather, it is more like the capacity of a balloon. A bit more capacity can usually be squeezed in, but the balloon (or route) becomes more stressed and resistant to any further increase. In this context, capacity restrained assignments are recommended where networks are congested.

The application of a capacity restraint does not imply that once a certain level of traffic is reached, no more traffic can be assigned, as the theoretical capacity may be exceeded albeit with resultant lower travel speeds. All the traffic derived from the demand forecasting will be distributed to the road network in one way or another, and usually on the basis of volume delay functions.

It is common for traffic assignment models to employ volume delay functions, which describe how the speed or travel time on a particular link will deteriorate as traffic builds up.[4]

A number of specific techniques are available for undertaking a capacity restrained assignment (such as equilibrium, stochastic, volume-averaging, incremental). Detailed information on these techniques is provided in the available literature.

The capacity restrained assignment generally begins by estimating the shortest path from each origin to all destinations (expressed in terms of the minimum generalised cost composed of some combination of travel time, travel distance and tolls). Trips for each origin–destination pair are then assigned to the links comprising the minimum path. As traffic volumes build up on a particular link, the speed on that link declines and the travel time increases making the link less attractive to any further traffic, leading to alternative paths to the same destination becoming more attractive.

The outputs from the highway traffic assignment are traffic volumes across the highway network, vehicle-kilometres and vehicle-hours of travel, trip length, peak period and daily traffic volumes, congested speeds and travel times and volume-to-capacity ratios. It is desirable that traffic assignment should be performed for peak periods, and off-peak periods with daily traffic volumes computed by combining the peak period and off-peak period traffic volumes. The output from the traffic assignment can be used in economic appraisal and congestion analyses.

Most commonly static assignment methods are used in strategic models. These assume network conditions and routeing does not vary within the modelled period. A dynamic assignment approach (Chiu et al, 2001) is more complex, representing variations in conditions and routeing choices within the modelled period. This allows the model to represent the formation of queues, and the metering of demand downstream from constrained junctions, providing a more detailed understanding of the network performance. In some cases the modelling also distinguishes lanes allowing for example an improved representation of effective network capacity and the implications for delays and impacts at merges and diverges where traffic may not be spread evenly between different lanes.

It is however common to represent more detailed aspects of network performance using microsimulation tools complement strategic models, as discussed further in Section 3.6.3.

Treatment of tolls

There is variation in the willingness of travellers to pay and, where there are choices between tolled and non-tolled routes, it is important for this to be represented. There also tend to be marked differences in the congestion or delay on the tolled and non-tolled roads, and differences in perception of time spent in queues can also influence route choice. A range of approaches has been applied to model demand for toll roads. Depending upon the needs of the study, one of these options would represent current best practice:

  • Explicitly represent the choice between use of the tolled road and use of other routes through a choice model e.g. logit.
  • Segmentation of demand between different value of time categories explicitly to represent the distribution of value of time
  • An assignment algorithm incorporating an explicit representation of the value of time distribution.

Further discussion of toll road patronage forecasting will be available in subsequent updates to the ATAP Guidelines.

3.6.2 Public transport assignment

Public transport assignment procedures predict the route choices for public transport trips on the basis of the different attributes of the public transport network. Some of the more critical attributes are:

  • Supply of public transport services as defined by the capacity of the public transport vehicle and its corresponding frequency. The public transport network consists of the route segments (links) and public transport stops (nodes) that form the public transport routes (lines).
  • The estimated cost of using public transport services is the average fare paid to take the trip.
  • The generalised impedance of travel by public transport is a function of the in-vehicle time, the time spent waiting, the time spent getting to a public transport stop, the time spent transferring from one route to another, comfort and convenience, public perception of the quality and reliability of each mode and the fare paid.
  • Some of the main outputs from the public transport assignment include public transport patronage and line or service loadings, boardings and alightings at stops, interchanging within and across modes, and network-wide indicators such as passenger-hours of travel and passenger-kilometres of travel. Other outputs often required include station entries and exits (subtly different from boarding and alighting).

For studies where public transport passenger demand forecasting is the focus, then it is highly desirable to separately model the mode of access to the public transport system, e.g. walking/cycling, park-and-ride and kiss-and-ride.

3.6.3 Traffic simulation methods

The management of a road network often requires the forecasting of the impacts of implementing various traffic management measures. The impact involves the road itself, the whole corridor and its abutting areas. These measures include, for example, signal coordination, high-occupancy vehicle (HOV) lanes, one-way systems, different types of intersection control (priority sign, signal or roundabout), signal priority, driver information systems and incident management. Apart from road vehicles, trams, light rail vehicles, pedestrians and cyclists can also be simulated.

Simulation techniques for traffic assignment can be broadly classified into the following five types:

  1. Macrosimulation: Assignment models at strategic level, part of the traditional four-step models and applied to large-scale areas. Typically, they are applied to time period O-D matrices, they use simplified representations of network links and nodes and make use of relatively simple volume-delay functions.
  2. Mesosimulation: Traffic models that operate at a less disaggregated level of detail than the microscopic models. They are based on traffic flow theory and apply analytical procedures that do not require random sampling from statistical distributions of input variables. They are particularly useful for area-wide traffic management and congestion management issues.
  3. Microsimulation: Simulation models that provide a very detailed view of the traffic by tracking movements of individual vehicles through the network and updating frequently (seconds) the position, speed and trajectory of each vehicle. They operate on the basis of traffic flow theory (e.g. car following theory, gap acceptance and lane changing behaviour), accounting for the behaviour and characteristics of the traffic participants (driver, pedestrian, cyclist). They require a detailed description of the network (design of links and nodes/intersections).
  4. Nanosimulation: The most refined level of traffic modelling, seeking to replicate the behaviour of individuals using different modes of travel. It is particularly concerned with modelling waiting times, interaction between individuals, etc. The model requires network description similar to micro-simulation, however enriched with data on pedestrian spaces and corridors. They can also be used in the design of transport terminals (such as railway stations) and access to buildings and other facilities.
  5. Hybrid modelling: Combination of mesoscopic and microscopic models in a single modelling framework. It allows the modeller to perform microscopic simulations within focus areas inside the full network mesoscopic simulation of the network.

In recent years, Intelligent Transport System (ITS) measures such as adaptive signal control algorithms, incident management strategies, active bus/tram priority and driver information systems have been introduced to freeways and arterial roads. These are complex traffic processes and traffic flow theories are often unable to accurately predict the impacts in terms of delay, queue length, travel times, fuel consumption and pollutant emissions. Computer models equipped with advanced graphical facilities have been developed in recent years to meet the needs of road managers.

Computer software has long been available to simulate traffic management processes amongst road authorities in Australia (see for example Cotterill et al. 1984; Tudge 1988). Past research also includes the development of car-following and lane changing algorithms for microsimulation (Gipps 1981, 1986), the review of eight small area traffic management models (Luk et al. 1983) and the comparison of macroscopic and microscopic simulations (Luk & Stewart 1984; Ting et al. 2004).

More recent research includes the assessment and further development of car-following and lane-changing algorithms (Hidas 2004, 2005; Panwai & Dia 2004). A key finding is that microscopic simulation models require careful calibration to produce meaningful results, especially in relation to lane changing behaviour in congested conditions.

3.7 Activity-based modelling

An alternative approach to the four step and five step models (from now on referred to as four step models for simplicity) are activity based models. Activity based models attempt to address some of the deficiencies inherent in four step models and introduce a disaggregate basis for estimating transport demand.

3.7.1 Deficiencies with four step models

The four step models provide strategic modelling frameworks that have been widely adopted by the transport modelling community throughout the world. The success of the four-step model in becoming the most commonly used form of strategic transport model can be attributed to:

  • the four (or five) steps being conceptually relatively easy to understand;
  • the ability to implement the models using off the shelf proprietary software without the need to be able to code in complex programming languages; and
  • the ability of these models to be able to appear to provide answers to the questions policy makers put to transport modellers, particular with respect to network performance

However, limitations and deficiencies of the four step process have long been recognised (see McNally and Recker, 1986; USDOT, 1997). The deficiencies of four step models can be summarised as:

  • not explicitly recognising that travel is a derived from people’s activity patterns and focusing on individual trips or tours, rather than reflecting the patterns of behaviour, and interactions within particular households;
  • presenting travel behaviour as an outcome of a clean choice process, rather than being defined by a range of complex constraints such as being influenced by household dynamics and social structures;
  • four step models inadequately specify the interrelationships of travel and activity between individuals and scheduling of activities in time and space.

3.7.2 Characteristics of activity base models

Activity based models attempt to address these deficiencies of four step models by using the household activity pattern as the basic unit of analysis. Activity-based models recognise that travel is derived from the demand for activity participation and therefore need to consider the interdependencies and constraints involved in scheduling activities. This allows activity based models to be able to address many transport policy questions that traditional four step models are unable to adequately model, particularly with respect to travel demand management policies that do not relate to the provision or removal of transport infrastructure.

Activity-based models apply functions to simulate which activities are conducted when, where, for how long, and with whom. These activity patterns help determine the travel choices available. Because of the complex nature of household activity patterns, activity based models lend themselves to be microsimulation in nature. This does not relate to the level of spatial disaggregation, rather it is an indication of whether the model is based on the individual decision making process, that is microscopic, or whether the model tries to simulate the overall transport pattern - macroscopic.

The development of activity based models can be distinguished by three different approaches:

  • constraint-based models attempt to determine the travel and activities that are feasible within particular space-time constraints; an early example of a constraint based model is PESASP (Lenntorp, 1976);
  • utility maximising models assume that individuals tried to maximise their utility of their daily activity schedule (Ben-Akiva, et al., 1996), and PCATS (Kitamura and Fujii, 1998); and
  • computation process models uses decision heuristics, an example being ALBATROSS (Arentze and Timmermans, 2000).

A common component of activity based models is a population synthesizer, which is used to create a synthetic population. Activity patterns and travel are determined at the disaggregate level and then aggregated up for analysis. A typical activity based model may contain the following features:

  • population synthesis
  • daily activity pattern formulation
  • tour formulation
  • time of day and mode choice

A consequence of this process is the need for computation resources that substantially exceed those generally used for four step models. Activity based models are also generally developed in an open source environment as the execution codes have yet to be implemented in a standard form to allow use in propriety applications for widespread use.

Activity based models are more common in the US than the rest of the world; however, even in the US they are still in the development stages and not yet standard practise. Where activity based models are used, the focus is generally on testing non infrastructure related policies such as road pricing and demand management strategies. An example of this is in the Netherlands where the activity based ALBATROSS model is used to test transport and land use policies that four step models are not able to adequately address, whereas, infrastructure projects are generally modelled with a tour based model.

3.7.3 Potential application in Australia

Activity based models are gaining some traction in the US and the Netherlands with proponents for activity based models arguing that activity based models address many of the deficiencies of four step models. However, the uptake of activity based models has not progressed at a rate to suggest they will replace the four step model as the most widely used strategic transport modelling tool in the near future.

A number of factors may be contributing to this slow uptake, chief amongst these could be the perceived increase in model complexity in moving from macroscopic four step models to microscopic activity based models. The current inability of activity based models to be implemented for large metropolitan areas without the need to use complex open sourced software, significant hardware costs and associated limits on the stability of model outputs is a barrier to widespread adoption.

It is also not entirely clear whether activity based models are superior to four step models for the appraisal of infrastructure projects. Infrastructure appraisal requires transport models to produce unique well converged solutions; the economic analysis is very sensitive to small changes in costs output by the transport models.

If an activity based model is considered for implementation, travel diaries in the form of activity are required rather than trips or stops in the traditional household travel survey. All activities that are conducted out of home would need to be collected. The more detailed the model, the higher the breakdown of these activities should be. The other dimensions of activities such as their location, time, companion (people who do activities together) and transport mode would also be required. Arentze et. al provides a detailed discussion on the data need for the development of activity based model.

3.8 Freight and commercial vehicle modelling

The focus of this volume is the travel demand modelling of person trips. The modelling of freight and commercial vehicle trips is also vitally important for planning and infrastructure planning purposes. Freight and commercial vehicle modelling is a complex field and hence requires a separate volume to cover adequately. The development of a volume focusing on freight and commercial vehicle modelling will commence later.

[1] The ‘gravity model’ in transport modelling is analogous to Newton’s law of gravity, as in both models the attraction between two bodies (or zones) is proportional to their masses (trip generation) and inversely proportional to their separation.

[2] A combination of time, distance, tolls and other out-of-pocket travel costs such as parking charges.

[3] More formally, the elasticity of demand with respect to a particular parameter is defined as the ratio of the proportionate change in demand to the proportionate change in the relevant parameter.

[4] Defined by the volume-to-capacity ratio.