Transport models should be designed to be suitable for the particular application(s) envisaged. Thus, the design should be sensitive to, for example, the context and the transport issues to be investigated. It should also allow for the availability of suitable travel data on which the transport model will be based and should consider such practical requirements as the turnaround time for a transport model run and the ease-of-use of the transport model.
That is to say, the objectives for the transport model should have a significant influence on its specification.
For example, city strategic transport models (such as four step models) are rarely appropriate for evaluating individual road and public transport projects. This is because they tend to be inaccurate in their representation of the travel on the transport networks in local areas of the city affected by the particular transport projects.
The reasons for this are threefold:
For reasons such as these, it is usually preferable for transport project appraisal to develop specialist public transport or road transport models focused on the particular project(s) which is (are) to be appraised. Although these transport project models are usually linked to the overarching city strategic transport model, they are separate and largely independent transport models which can be designed to focus on the particular needs of the project application.
These transport project models are more detailed in the local area of the transport project, and will include additional technical features needed for accurate project forecasts (such as detailed modelling of road junctions or a representation of rail passenger crowding). They will also make use of a wider range of travel survey data, which will improve the accuracy of the representation of travel patterns in the area local to the transport project.
For a discussion of the roles of the different types of transport models, see my section of the Monash University textbook on Transport Engineering and Management (2003).
Over the past 30 years or so, these sorts of arguments have also led to a greater use of "marginal" or "incremental" multimode transport models (commonly referred to as "pivot point" techniques). In my experience there have been two strong reasons for this.
In what follows I discuss two such multimodal projects to bring out some of the benefits of incremental modelling and how the transport models themselves were developed.
The concept of the Long Distance Travel Model for the Netherlands developed from the Dutch Government's Transport Structure Plan which identified a number of potential high speed rail routes in the Netherlands which could connect with a Western Europe high speed rail network.
Travel demand models operating at a national scale which had been developed using the fairly conventional urban methods in use at that time had been shown to give reasonable forecasts of car travel but unsatisfactory predictions of interurban rail travel. The explanation of these failings was found when surveys showed that train travel in the Netherlands comprised mainly journeys over longer distances, with characteristics very different from shorter, urban trips. [Around that time, similar model failings were exposed in the UK Regional Highway Traffic Model project.] This caused the Ministry of Transport and Public Works to take a special interest in studies of long distance travel which in turn led to the commissioning of a project with The MVA Consultancy to develop a model of long distance travel in the Netherlands (long distance travel being defined as trips over 40 kms).
A particular motivation was the need to appraise a proposed long distance railway line, the Zuiderzeelijn, running north east from Amsterdam across the new Flevoland Polder towards Groningen.
There were two phases of data collection.
The first was a programme of social research involving in-depth interviews with 200 people who had made recent long distance trips. This was used to inform the designs of the transport model and of the main intercept travel surveys.
The second phase was a programme of intercept data collection in which rail, car and bus users were intercepted in three major long distance travel corridors, one of which was the corridor in which the proposed Zuiderzeelijn would operate.
The LDTM was developed as a fully disaggregate multimodal transport model. It adopted a "marginal model" structure in which the travel forecasts for a particular mode of transport were obtained by estimating the changes to an observed base pattern of travel for that mode (as developed from the intercept surveys). This approach meant that the travel demand forecasts were not dependent on the accurate synthesis of base year mode-specific travel patterns, which had proved impossible in the earlier national studies of long distance travel.
Models were statistically estimated for mode choice and trip frequency from the intercept survey data and combined with other models, of travel growth, the time of travel and of the road and public transport networks.
LDTM was used to produce forecasts for the proposed new railway, the Zuiderzeelijn, the survey data base being that collected in the development of the model.
Subsequently, the Netherlands joined a group of European countries (Belgium, France and Germany) to study the feasibility of a proposed new high speed rail route from Paris via Brussels to Cologne with an additional link between Brussels and Amsterdam. The Dutch contribution to the passenger forecasts for this high speed rail route was based on an adaptation of LDTM.
The application of LDTM to these forecasts and the adaptations required are described in Passenger Forecasts for the Paris-Brussels-Cologne-Amsterdam High Speed Rail Line Study (authors: Ashley, Kooman and van der Star).
The key point to be made is that this application was made easier by the incremental approach, because a sophisticated model framework existed in which could readily be incorporated both new travel survey data and extensions and adaptations to the model specification. Specifically, extensive new travel intercept data was collected on the relevant international routes by road, rail and air between the Netherlands and the other three countries and incorporated in the LDTM; in addition, the LDTM network models were extended to cover the international linkages and the mode choice model was enhanced to allow for air travel (using stated preference surveys).
I was involved in modelling surface access to the three main London airports: Heathrow, Gatwick and Stansted over more than 20 years, between 1986 and 2008. This started with a study for the UK government of improved surface access links to central London which eventually led to the provision of the Heathrow Express service into the airport. Many more studies followed including those associated with the development of Heathrow Terminal 5 and, most recently, the expansion of Stansted Airport.
The model system covered both air passengers and airport employees, but it is the air passenger model to which I refer here. In the early studies this was called the Heathrow Surface Access Model (HSAM) and in the more recent studies, the London Airports Surface Access Model (LASAM).
HSAM/LASAM are mode share models which predict the annual average surface access mode shares of air passengers at the London airports. The model system developed for the Stansted public inquiry is described in The Suite of Surface Access Models Supporting the Expansion Plans for Stansted Airport (authors: Ashley and Brown).
Using disaggregate techniques, these models were estimated on the UK Civil Aviation Authority’s annual surveys of air passengers at the three main London airports (Heathrow, Gatwick and Stansted).
The transport model was implemented separately for each airport as an aggregate, incremental model (at the zonal level), with appropriate adjustments. The theory of the incremental modal choice model is explained in The Nested Incremental Model: Theory and Application to Modal Choice (authors: Bates, Hyman and Ashley).
The data base available was large, in excess of 100,000 interviews with air passengers, providing a suitable basis for estimating the observed travel patterns by each mode of transport to each of the airports, although some smoothing of the data was used where data samples were sparse.
There were some significant advantages to this approach: