Land-Use / Transport Interaction Models
TAG Unit 3.1.3

June 2005

• in Section 2, the general principles of land-use modelling are set out; and
• in Section 3, the different kinds of land-use model are introduced.

Some studies may require the use of a land-use/transport interaction model rather than simply a transport model. In simple terms, the transport model requires inputs of land-use which have been forecast exogenously, whereas land-use/transport interaction models generate their own forecasts of land-use dependent on input land-use policies and the changes in accessibility brought about by conditions on the transport system. As land-use models are not in wide-spread use, the principles of land-use models are explained in some detail.

2.2 The Meaning of 'Land-Use'

The term 'land-use' is used throughout this report to mean a range of human activities, the state of the built environment, and also to some aspects of the natural environment.

• land-using activities and the interactions between them generate the demands for transport;
• those activities and interactions are to a greater or lesser extent influenced by the availability of transport; and
• the linkages between transport and activities may be important to the appraisal of transport strategies - especially when trying to consider whether the transport system is providing the kinds of accessibilities that activities (i.e. people and businesses) require, rather than simply providing mobility.

• the population, as individuals and as households;
• firms and other productive organisations; and
• government.

Figure 2.1 Actors and Markets in Land-Use/Transport Interaction Models

• property developers,
• transport infrastructure providers, and
• transport service providers (e.g. public transport operators),
  

which may be special cases either of firms, or of government activity, or both.

The term 'land-use' includes all of the elements and interactions in Figure 2.1 outside the area labelled 'transport', except for those particular effects which we define as 'environmental'.

• in property,
• labour, and
• goods and services.

Through these interactions, changes in transport may have indirect impacts on people or businesses who have no direct interest in the transport change at all. It may therefore be necessary to consider not only predicting the land-use consequences of transport change, but also the implications for appraisal of the way in which the influence of transport is passed on through the interactions of different actors.

It is important to recognise that the 'land-use' system is never static, and that 'transport' is only one of the factors that influence how it changes. The treatment of all the other factors - such as demographics, the workings of the development process, etc. - are among the things which distinguish the different approaches to land-use modelling reviewed below.

• the land-use impacts of a transport change may extend far beyond the spatial scope of the transport proposal itself - they can extend at least as far as the area in which the transport change affects accessibility, and secondary effects may extend further;
• a great deal of locational change takes place through changing occupation of existing buildings, with changes in either the density or the nature of the occupation (for example, one type of business replacing another, or retired persons occupying housing previously occupied by families with children);
• the value of property is an important influence on its occupation; if improvements in transport increase the demand for space in a particular location, the resulting increases in rents may affect households or businesses who have no direct interest in the transport change itself; and
• it follows from the above points (a) that in many cases changes in composition are likely to be more significant than changes in totals - for example, changes in provision for commuter travel may have a significant impact on where the working population and its dependants live, but a much smaller impact on the distribution of the total population (as households without workers move into the areas that the workers are leaving); and (b) that significant land-use effects may occur within the market for existing property, with no new development and no formal change of use, and therefore beyond the control of the planning system.

It should also be noted that 'regeneration', 'socio-economic impacts' and so on are all particular cases of what are here referred to as land-use effects.

The 'land-use' components of 'land-use/transport models' cover varying proportions of 'land-use' as defined above. In most cases their representation of the physical use of land is only a small part of the overall model. In some cases, the physical use of land is not considered at all.

2.3.2

A critical aspect in reviewing land-use models is not only to consider what categories of activities they represent, but what kinds of responses of those activities can be predicted - for example, whether households can choose not only to relocate, but also to change the type or size of dwelling they occupy. To begin this discussion, Figure 2.1 has been expanded so as to identify, in Figure 2.2, the main types of decisions made by the different categories of 'actors'. An additional category of actors has been added, that of 'investors', in the top right. In the conceptual model, this includes all those investors who may invest in the area under consideration, many of whom are resident outside the area itself.

For clarity, no attempt has been made to show within the diagram that many individuals are actors in more than one category - for example, self-employed persons are producers as well as residents, and many residents are also investors. Note also that one of the most important 'actors', government of all levels, is omitted, even though its intervention through regulation, taxation and investment is an actual or potential influence on almost all the decisions considered. Much of the development of operational models has been led by the need to consider the impact of such interventions, given the behaviour of all the other actors involved. It is important to keep in mind that in land-use modelling, the location of activities (and in many cases the location of the development they occupy) are outputs of the model, and that the models take a description of 'planning policy' as input; this contrasts with the conventional 'planning data' used in transport-only models, which corresponds with the outputs (e.g. population and jobs by zone) of land-use models.

• the arrowheads show the delivery of a factor, good or service; and
• payment for that factor, good or service goes in the opposite direction to the arrow.

Figure 2.2 Actors and Markets in Land-Use/Transport Interaction Models: An Expansion

Information also flows in both directions along each of the relationships indicated by arrows. This represents the often very partial information which people and firms obtain from the interactions with the market. Other information is obtained in other ways, which may themselves involve purchasing goods and services (e.g. market research reports, special surveys, newspapers with job and property advertisements, etc).

• the financial market(s);
• property markets;
• labour markets;
• product markets (including both goods and services); and
• transport markets.

• exported;
• consumed by the government; or
• used in fixed capital formation (the arrow from product markets to the 'invest' action of producers); and also that they may be;
• supplied by imports to the economy under consideration as well as by local production.

• transport demands associated with product markets, that is, with the delivery of goods and services (through the movement of goods and persons, including consumers going to purchase goods or services) to intermediate or final consumers;
• transport demands associated with labour markets - mainly the movement of persons travelling to work;
• other travel demands associated with the activities of producers - these represent all business demands, mainly for passenger travel, not directly associated with trade in goods or services (e.g. travel to conferences, to internal company meetings, to meetings with regulatory bodies, etc.);
• residents' travel demands other than travel to work or to obtain goods and services, i.e. all other personal travel; and
• transport demands associated with transport supply itself (e.g. the significant proportion of rail freight which is generated by maintenance and renewal of the railway itself).

• where to locate the business unit;
• investment in the unit - how much to invest, in what equipment;
• recruitment - what categories of staff to employ, how many, for what hours, at what wage rates, etc;
• purchasing - what intermediate goods and services to purchase, from whom;
• production - how much of what to make and when; and
• marketing - which markets to try to sell in, what to do to achieve this, etc.
  

Many decisions, particularly major ones, will of course deal simultaneously with most or all of these areas.

• where to locate (and hence what land and floorspace to occupy);
• training - what (if anything) to do to obtain/maintain employable skills;
• work - whether or not to work, for whom, doing what, when, etc;
• purchasing - how to spend (or save - note link to investors) income derived from work or other sources; and
• other activities - everything else.
  

Note that the first three determine each person's involvement or otherwise in the labour market, and hence collectively the 'labour supply', whilst 'labour demand' is determined by the location and recruitment decisions of producers.

• technological progress as an 'exogenous' influence on producers (in the sense that even if firms are technological leaders and innovators in their particular fields, they are strongly influenced by the development of technology in other aspects of the economy);
• natural demographic processes (ageing) and social effects (marriage/cohabitation, separation) on residents and their grouping into households; and
• network effects (congestion) in the transport system (as distinct from the deliberate responses of transport operators and suppliers).

• some form of spatial representation of producers, residents, and transport supply (not necessarily traced back to transport suppliers);
• links from the transport markets to the activities and markets which use transport, such that changes in transport have at least some impact on some decisions or responses of producers and residents; and
• scope for links from producers and residents, and/or from labour and product markets, to transport markets, as the main or only process by which transport demands are derived.

• to apply a simple model which predicts the land-use impact of transport change on the assumed planning data, but does not include the feedback from that land-use impact back to transport; or
• to apply a more complex model which includes both linkages, i.e. a full land-use/transport interaction model.
  Note that, in the former simple model case, some other part of the modelling system will need to convert the assumed planning data into future transport demand; in the case of a full land-use/transport interaction model, this will be a central part of the overall model system.

The DSC/ME&P Report to SACTRA discusses the scope of various models, including sketching out their coverage of actors and markets as defined in Figure 2.2; the significance of this is considered further in Section 3.

The range of models currently available is considered below, along with the extent to which they address the scope for modelling defined above. At this point in the discussion, it is useful to introduce one major distinction which allows the available models to be classified in two broad approaches.

The discussion around Figure 2.1 and Figure 2.2 has already mentioned that the subject of 'land-use' includes both the location of activities (and various aspects of their behaviour in those locations) and the economic interactions between activities in the various markets. These economic interactions - such as the flow of labour from homes to workplaces, or of goods from producers to consumers - are not generally identical with transport demands, but are clearly closely related to them. There is equally clearly a close relationship or identity, in many cases, between the measures of economic interaction and certain measures of activity location: for example, the row or 'home' totals of a matrix of labour (measured in workers) flowing from homes to workplaces must equal the number of working residents living in each zone, whilst the column totals of that matrix must equal the number of filled jobs in each zone.

Models can be classified according to how they link the location of activities and the spatial interactions between activities. This classification is important both to understanding the thinking behind different models and to practical questions of how they are implemented and used to appraise policies.

One approach takes the interactions between activities as the key variables. These are predicted, and then the location of activities is calculated from the total levels of interaction. For example, the number of households living in a zone is found by predicting the number of workers commuting from that zone to each possible workplace, finding the total workers resident, and then factoring from workers to households. The patterns of interaction are also factored, from persons to trips, to give transport demand matrices. This approach may be called the "interaction-location" or IL approach, since the central feature is that the predicted interactions determine the location of activities. Such land-use/transport models can also be called "integrated" models, since the distribution of transport demand is wholly predicted within the land-use model. This means that the land-use and transport components of the overall model system cannot be separated.

The alternative approach first predicts the location of land-using activities, and then models the interactions between those located activities. This can obviously be called the "location-interaction" (LI) approach. This allows the number and location of the different kinds of activities to be determined by separate sub-models. These can consider any appropriate influences, but include measures of zonal accessibility, which reflect the scope for interactions from each zone. Hence, for example, a sub-model for residential location will include measures of accessibility to work and to other destinations. The interactions between activities are then controlled by the location of activities. These interactions may be modelled in economic terms, or may be predicted directly in terms of travel demand.

This LI approach leaves the distribution of transport demand at least partly to the transport model. The overall model therefore consists of a complete transport model linked to a land-use model, rather than part of the transport model being embedded within the land-use model. The approach can therefore be described as "linked", in contrast to the "integrated" approach described above.

Various points about these alternative approaches should be noted.

Firstly, IL or integrated models tend to be defined in terms of finding the equilibrium location and interaction of the different activities considered, given certain fixed variables such as a "basic" or exporting sector of employment and the supply of land or floorspace. This is necessary because of the way in which the number and location of activities is built up from their interactions with other activities. For example, this approach generally requires that households are "generated" by the demand for their labour, and that demand depends in part on households' demands for services. This linkage has to be run to equilibrium, otherwise households and jobs will disappear from the system.

In contrast, the LI or linked approach need not have any equilibrium between the location and number of different activities - it can for example readily predict an increasing supply of labour in an area of decreasing demand, and the resulting unemployment.

Secondly, IL models by definition predict matrices of interactions which can be converted into matrices of the demand for transport. This may be useful in circumstances where observed transport demand data is unavailable or where a synthetic matrix is needed as input to a matrix-refinement process.

• to develop a land-use/transport model where a suitable transport model already exists, by adding an appropriate land-use element; and
• to carry out transport-only tests, and hence to produce transport-only calculations of benefit which are currently required as part of the appraisal process.

It is probably helpful to split the issue of appraisal in relation to land-use/transport modelling into two subjects: first, the appraisal of transport strategies or plans, with land-use policy held constants, and secondly, the appraisal of alternative land-use policies, alone or in combination with alternative transport strategies or plans. Note that, for the reasons already explained, holding land-use policy constant does not mean that the land-use patterns will remain constant - they may be changed by the effects of the transport strategy or plan.

The first point to note is that it is currently not possible to conduct a cost/benefit analysis in which land-use changes feed through into travel demand changes. The reason is that, at present, the way in which land-use responses and transport responses are represented mathematically in land-use/transport interaction models are not sufficiently consistent to allow the calculations to be undertaken in a manner which accords with the theory on which transport cost/benefit is currently based.

The economist's conventional view of the land-use impacts of transport change has been that such impacts change the distribution of costs and benefits - for example, transport benefits initially enjoyed by travellers may be captured by real estate owners through increasing rents - but that they do not modify the total net value resulting. This view would imply that it is not necessary to assess the benefits associated with the land-use impacts at all, because they are simply transformations of the benefits which can be estimated on the basis of a transport-only analysis. There are at least three objections to this view.

The first is that the distribution of benefits is often of concern, both spatially and socially. Most governments have policies which are intended (for example) to redistribute jobs to high unemployment areas, and transport investments which support such policies should be regarded as more beneficial than those which work against them.

The second is that the view that land-use impacts transform and redistribute transport benefits has been shown to be valid only under conditions of perfect competition (Jara-Díaz, 1986). More recent work (Martínez and Araya, 1998) has shown how unrealistic these conditions are, and has started to show how much the measures of benefit are modified by land-use effects. Whilst it appears highly significant, there is much more work to be done both on the issues it raises about how benefits should be evaluated, and if appropriate to implement suitable methods of benefit calculation within other land-use models.

The third, which is perhaps a less formal view of the second, is that if the costs and benefits of a transport scheme change as one expands the scope of the transport analysis, it is implausible that the costs and benefits should not differ further if the analysis is extended into 'land-use' effects. For example, the appraisal of a major motorway project will produce one result if it is based upon a fixed matrix of person-trips by road, but a different result if modal choice is taken into account and public transport operator response to changing demand is taken into account (e.g. if transfer from rail to road will lead to a decline in rail services). It is hard to see why further extension to include location and development effects would not lead to further modifications of benefit, especially when one notes that the location effects may be influenced by environmental externalities as well as by (variables derived from) the generalised cost of transport.

At present, there appear to be two approaches to appraisal in land-use/transport modelling practice.

One effectively ignores the issues identified above, and carries out a relatively conventional transport-only calculation of benefits (based primarily on time and money savings) by testing the alternative transport strategies with land-use held constant. This could be extended by carrying out the test under both the reference case land-use pattern and the modified land-use pattern resulting from the land-use impacts of the strategy being tested; this would show whether benefits increased or decreased as land-use responded. Neither of these sets of calculations would yield the net benefit arising from the combined effect of the transport strategy and the resulting land use response nor would it tell us how the land-use effects would redistribute benefits.

However, it is possible to examine the predicted land-use effects and to include separately in the appraisal (as envisaged in the new Appraisal Summary Table, see Transport Appraisal and the New Green Book (TAG Unit 2.7)) any impacts which are identified as being particularly desirable or undesirable. Desirable impacts would include regeneration (however defined, e.g. new development, new jobs, or reduced unemployment) in areas where that is a policy objective; undesirable impacts could include very much the same effects in areas where they are considered undesirable (e.g. increased demand for housing and associated pressure for development in National Parks or AONBs).

At the other extreme, at least one modelling package attempts to carry out a comprehensive appraisal of benefits in the land-use system. This does respect the requirement to take land-use effects into account, as outlined above. This, of course, produces measures of benefit different from those in transport-only analysis; it should include the benefit (or disbenefit) that households or firms obtain from paying different levels of rent, from living at lower or higher densities, from being in different (e.g. more or less attractive) locations, and so on. This clearly goes well beyond a conventional transport cost/benefit analysis,; however desirable such an extended analysis is, it may come up against the institutional or administrative problems because it is unfamiliar and difficult to relate to the analyses from more conventional models.

Such analysis also raises more technical complications in terms of dealing with the lagged responses to transport in most land-use models. It is no coincidence that the most advanced application of the analysis of both land-use and transport benefits within a land-use model is based upon a system in which the land-use and transport systems are calculated so as to be in complete equilibrium with each other. The more sophisticated the dynamics of the land-use modelling, the more complicated it will be to establish sensible measures of benefit calculated within the land-use system.

There is also an issue of the assessment of environmental effects. In some land-use models, residents (and potentially firms) are influenced in their location decisions by the environmental impacts of transport. Negative transport impacts (e.g. increases in noise and in local air pollution) would decrease the willingness to pay to live in the locations affected, and would generate disbenefits (e.g. to the owners of property in those locations). This might start to duplicate environmental impacts which have conventionally been considered as separate parts of the overall appraisal process.

A major attraction of the comprehensive appraisal of benefits (including the benefits derived from transport) in a land-use model is that such an approach should in principle be able to carry out a consistent appraisal of any combination of land-use and transport elements. This needs to be considered not only as an extension of transport strategy appraisal, but also in terms of its possible role in the land-use planning process.

The idea of a consistent, combined appraisal of land-use and transport choices has a theoretical appeal, and should help to ensure that the wider objectives of land-use planning are not made subordinate to the narrower objectives of transport planning. However, it is beyond the scope of the current practice, and therefore this Guidance, to investigate the complexities of assessing costs and benefits in this area. It is, however, appropriate to note that existing land-use models can provide a wide range of indicators (not just transport indicators) about the impact of alternative land-use strategies, alone or in combination with transport strategies. These indicators are the kind of information expected by current approaches to assessment in both fields of planning, under headings of 'regeneration' or 'socio-economic impacts' as well as 'land-use' itself.

Although land-use/transport interaction (LUTI) models have been in use for many years now, their use in transport strategy development in this country has been, until now, quite limited. As a consequence, there is much less familiarity among practitioners about these models than with the various forms of transport model. Other information on land-use/transport interaction models is in Transport Models (TAG Unit 3.1.2).

The various kinds of LUTI models are classified in Figure 3.3. The first layer of the tree, starting from the top, separates out a group of models whose purpose is to optimise urban systems rather than to predict their behaviour. Such models are intended as tools which can find a 'design' to optimise a particular function, and are therefore quite distinct from the majority of models which respond to a 'design' input by the user. These optimising models may be informative for research and long-term planning, but in general they require a substantial model development effort in order to link them to the practical planning problems of individual cities or regions. Accordingly, they are not considered further here.

Figure 3.3 Classification of Models

The second layer of the tree distinguishes between 'static' and 'dynamic' models. Static models represent a single point in time, whereas dynamic models run for a series of time periods, with transport changes generally taking one or more such periods to have an impact on land-use. Much of the early work in land-use modelling consisted of static models which attempted to predict the location of certain variables taking other simultaneous variables as given (see, in particular, Lowry, 1964, and the whole range of Lowry-inspired models considered in Batty, 1976). Such models obviously cannot represent in any 'realistic' way the processes of urban change which, by their nature, take time to react to any changing situation. For this reason, static models had ceased to represent the state-of-the-art by the time the ISGLUTI project began around 1980 (see note on ISGLUTI, following the references at the end of the TAG Unit). Static models have, however, retained some relevance to cases where a dynamic land-use/transport model is unaffordable.

• models based originally upon the analogies with statistical mechanics ("entropy") pioneered by Alan Wilson in the 1970s;
• models based primarily upon the integration into a spatial (multizonal) form of separately developed (and often non-spatial) economic models; and
• models based primarily upon representation of the different processes affecting the different types of activities considered.

The range of available models within each of the above groups is now outlined, concentrating on those developed or used in Europe and likely to be available to the present studies. A similar review on a wider basis, including more non-European and research-oriented examples, can be found in the DSC/ME&P Report to SACTRA.

• as a means of adding a land-use impact dimension to existing transport models, without embarking on the extra work needed to create a dynamic model; and/or
• because the static model represents an equilibrium state which is of interest in itself.

• models which estimate the pattern of land-use given one set of transport inputs; and
• models which estimate changes in land-use given two sets of transport inputs.

The Swedish IMREL model is representative of the single-input approach, whilst DSCMOD is representative of the two-input approach. DSCMOD has been developed by DSC since 1990 for the practical objective of adding a land-use dimension to what would otherwise be transport-only studies. IMREL (developed by Anderstig and Mattsson in Sweden, 1991 and 1992) has been used for both research and planning purposes. These models are described in more detail in Annex A of the DSC/ME&P Report to SACTRA. Many other static models were developed in the 1960s and 1970s for specific studies.

All of these models are linked to separate and usually pre-existing transport models. IMREL estimates equilibrium patterns of land-use corresponding with the accessibilities output by the transport model. DSCMOD, in contrast, assumes that the 'base case' land-use forecast is in equilibrium with the 'base case' transport strategy, and calculates changes in land-use from the accessibilities produced by alternative transport strategies. In DSCMOD, these accessibility changes may be the only influence on location choice, or may be combined in a more complex mechanism with floorspace constraints and market clearing using rent adjustments.

These models are generally urban models. However, a regional employment version of DSCMOD (Simmonds, 1992; Simmonds and Jenkinson, 1993) has been developed which represents only employment and uses a measure of economic potential (accessibility factored by zonal employment) to relocate jobs.

The main UK example of an entropy-based model was the LILT (Leeds Integrated Land-Use Transport) package (Mackett, 1979 and 1983), which, when developed in the late 1970s, was the most substantial land-use/transport model application for a British city. It is understood, however, that this package is not available for new applications.

MEPLAN (Echenique et al, 1990) and TRANUS (de la Barra, 1989) are both commercial packages developed from a set of models devised at the Martin Centre at the University of Cambridge⁽¹⁾ . Both MEPLAN and TRANUS have been applied in policy and research studies both in the UK and abroad since the 1980s. Each package includes both a land use model and a multi-modal transport model, and is usually implemented as a quasi-dynamic model. There are many similarities in the broad approach adopted by the two packages.

MEPLAN and TRANUS are the key examples of interaction-location models (see 2.29). The interactions ('economic trades') between activities are determined by input-output analysis, and these interactions are used to derive the demand for transport. Location choices, transport mode choices and assignment are determined in a multi-level choice structure based on random utility theory. The location behaviour of households, firms and property developers is based on competitive markets, with incomes and rents determined endogenously in each time period.

A new package, MENTOR, is currently being tested. MENTOR is a land use package that can be interfaced to existing transport models. It builds on the theoretical structures of MEPLAN but operates on a more detailed segmentation of activities and is designed to be more straightforward to set up and calibrate. It retains the key characteristic that the distribution of transport demand is explicitly derived from the interactions modelled within the land-use model.

The MEPLAN package, its application to LASER (a model of London and the South East, which focuses on residential location and the journey to work, shopping and schools) and EUNET (a model of the Transpennine corridor, which places emphasis on industrial location and the movement of freight), and the MENTOR package are all described further in Annex C of the DSC/ME&P Report to SACTRA (DSC and ME&P, 1999). Information on the TRANUS package, including its application in Swindon (which gives particular emphasis to the consumption of energy and generation of polluting emissions) may be found on the internet at http://www.modelistica.com.

Activity based models are defined by their focus on the different processes of change which affect activities and the spaces they occupy; they are location-interaction models, typically characterised by more detailed segmentation of activities, and more elaborate treatment of both the decision to move and location choice. In contrast to other models, they do not relocate all activities in a time period, but separate the decision to move (which will affect only a proportion of total activities) and the search for a new location. These models also represent demographic change in more detail than any of the models so far considered.

The best-developed model of this type is IRPUD (Wegener, 1982), a model of Dortmund (Germany) developed for research purposes over a long period. The one UK example is the DELTA package, which has been developed by DSC since 1994 (see Simmonds and Still, 1998; Simmonds, forthcoming). DELTA has been applied to Edinburgh and to Greater Manchester, and in an extended regional form (see below) to the Trans-Pennine region. A rather similar model, URBANSIM, is currently being applied in the USA to Eugene/Springfield (Oregon) and is to be applied to the Salt Lake City region.

These models are designed to be linked to transport models developed in separate packages⁽²⁾ . Each consists of a number of sub-models representing different processes of change; in the DELTA case, these are physical development, improvement or decline in area quality, car ownership, demographic and economic change, location and the property market, and employment status. One of the characteristics of the focus on processes of change is that the design and calibration of the model draw much more upon other aspects of urban research (in economics, geography, sociology, etc) rather than drawing purely upon other modelling work. Another characteristic is that different processes are likely to predominate at different spatial scales.

A regional version of DELTA has recently been developed and has been applied to the whole of the Trans-Pennine Corridor. This version retains the original processes of urban change within each of the conurbations and other areas, but additional processes of migration and of regional economic change are added to represent the demographic and economic interactions between them. This contrasts with the Martin Centre of representing different scales, from city to continent, by modelling different variables within the same spatial-economic framework.

More details of DELTA can be found in Annex E of the DSC/ME&P Report to SACTRA.

• what interactions between supply and demand are represented in the transport model;
• what information is passed from the transport model to the land-use model;
• what impacts changes in transport have within the land-use model (and whether these are immediate or lagged); and
• what information is passed back from the land-use model to the transport model.

These points represent a useful checklist for gaining an appreciation of any particular model. Another important aspect of understanding is to know which effects within the land-use model (out of the range discussed around Figure 2.2) are explicitly represented as decisions of particular kinds of 'economic actors' (households, firms, etc), or as other appropriate and explicit processes, and which are represented only implicitly by fixed relationships or as being determined by other decisions. (As an example of an effect determined by other decisions, service employment in the Martin Centre models is not modelled as a decision by the retail sector, but is calculated wholly as a consequence of consumers' decisions on where to shop.) Differences in representation cannot generally be described as 'right' or 'wrong', but particular approaches may well be 'appropriate' or 'inappropriate' to particular studies: the relationship between economic change and population change, and between both of those and the development process, are points which should be considered carefully.

One way of looking at these points is to consider the ways in which different models represent the markets, in labour, in goods and services, and in property. Summaries of the representation of decisions and of the resulting treatment of non-transport markets in a number of current models can again be found in Chapter 4 of the DSC/ME&P Report to SACTRA.

It must, however, be noted at this point that the modelling software in use consists of 'packages' which generally offer considerable scope for different applications within one broad approach. The detailed representation of response to transport in non-transport markets could be significantly different in future application of these packages. There is scope for fine-tuning to the requirements of particular studies, though the constraints on this should also be noted: above all, that the modelling of choices in the 'land-use' system, just like the modelling of choices in transport, is only valid if the set of possible choices is correctly specified. This makes it difficult, and at present largely impossible, to build a meaningful land-use/transport model for a small area around a scheme, or for a one-dimensional corridor between two places. This difficulty also applies to modelling the distribution of travel, and has already been identified in the Fearon Report.

The land-use/transport models considered in this section all include or require an operational transport model. It is therefore appropriate to consider under 'data requirements' only those requirements which are additional to those for transport modelling, i.e. the extra data needed for land-use modelling. Since some of the land-use models include functions which elsewhere are left to the transport model, this division is not exact; these points are noted where applicable in what follows.

• data required to implement the model, i.e. the variables which have to be introduced in order to make the model represent the chosen city or region, and which are either direct inputs to the working model or are automatically reproduced by the working model (see paragraphs 3.7.4 to 3.7.6); and
• additional data or information required to calibrate the model to reproduce the behaviour of the chosen system or the processes at work within it (see paragraphs 3.7.7 to 3.7.10).

• the range of inputs which can be used to specify future scenarios (see paragraphs 3.7.11 to 3.7.12), and
• the range of policies that can be tested (see paragraphs 3.7.13 to 3.7.14).

The following sections deal with these in turn. It should be noted at the outset that only a broad and general description of requirements and possibilities can be given; the details will depend on the design of each particular model application, and in nearly all cases the model implementation process can be adapted to the availability or otherwise of particular types of information.

Data requirements for model implementation. In general, the requirements for the implementation category of data are quite firm - there must be one number for each variable in each zone - but at least in the early stages of model implementation there is a lot of scope for choice in the definition of variables (e.g. how many household types, how many employment categories). In contrast, the requirements for calibration are much less precise - although some of the packages have automated calibration routines which require particular inputs, these are not the only way of arriving at the eventual coefficient values. The data requirements are summarised in Table 3.1.

Table 3.1: Implementation Data for Land-Use/Transport Interaction Models

Variable	Static Models		Dynamic Models
	IMREL etc	DSCMOD	Martin Centre	DELTA
Households/ population	Few categories to reproduce	Few categories input as base situation	Few categories to reproduce	Few or many categories inputs for base year and earlier
Employment (status of residents)	Not applicable	Not applicable	Not applicable	Required as input (may be reflected in household categorisation)
Employment (by workplace)	Few categories to reproduce	Few categories input as base situation	Few categories split into exogenous component, input, and endogenous, to reproduce	Few or many categories inputs for base year and earlier
Floorspace by type	Required for base and alternative situations	Optional	Required for base situation	Required for base situation
Rents	Required for base situation	Optional for base situation	To reproduce by calibration in base situation	Required input for base situation and earlier
Household incomes	not used	Optional for base situation	To reproduce by calibration in base situation	Required input for base situation and earlier
Matrices of labour to work	(in transport model)	(in transport model)	To reproduce by calibration in base situation	Required input for large study areas; also in transport model
Matrices of goods and/or services to consumers	(possibly implicit in transport model)	(possibly implicit in transport model)	To reproduce by calibration in base situation	(possibly implicit in transport model; explicit in regional versions)
Development under construction in base year	see floorspace, above	see floorspace, above	Not required	Input

• at the urban level the DELTA approach does not consider the pattern of trade in goods and services, only in labour;
• the DELTA approach generally requires rather more information about previous years (in line with its lesser assumptions of equilibrium); and
• a number of variables have to be reproduced by calibration in the Martin Centre models but are simply input to DELTA.

• it may mean that the calibration is optimised to reproduce all the cells of the matrix as well as possible, or simply that some characteristic of the matrix (such as the average travel to work distance) is reproduced, and that in the latter case the matrix itself may not be used at all (if for example average travel to work distance is obtained from a household survey); and
• data which is input may itself be synthesised, and especially in the case of matrices will always involve some element of synthesis. (In general there is decreasing difference between methods which synthesise their base situation but are adjusted (e.g. by residual disutility methods) so as to reproduce all the confidently-known features of the observed base data, and those which are intended to take the observed base data as input but require a pre-model synthesis of those cells which are not known with confidence.)

Approaches to calibration. It is worth emphasising that the design of DELTA does not envisage that all of these behavioural coefficients should be estimated on local data; it is seen as a positive advantage that they should be based on wider research, supplemented by local experience. The Martin Centre approach, in contrast, places much more emphasis on reproducing many aspects of the initial situation; it therefore lends itself to a rather more statistical form of calibration based on that reproduction in terms of simultaneous relationships, but it involves relatively few coefficients that relate to recognisable processes of change over time.

Approaches to calibration: Martin Centre Models. Table 3.2 (based on Hunt, 1994, and Hunt and McMillan, 1995) attempts to identify the coefficients which have to be set up in the cross-sectional elements of a typical Martin Centre-type model. All these are used in the "reproduced by calibration" elements of the database.

Table 3.2: Coefficients of Typical Martin Centre Models

• exogenous employment by sector:
• exogenous households⁽³⁾ by socio-economic group; and
• floorspace by category.

The increments (positive and negative) in these are exogenous to the model.

Approaches to calibration: DELTA. Table 3.3 lists the more purely behavioural coefficients of DELTA (as implemented for the Greater Manchester Strategy Planning Model (GMSPM) - see the Fearon report for further details). All of these are calibrated primarily on the basis of previous national or international research, supplemented where necessary with professional judgement and where appropriate with adjustment to the scale of the particular model. Note that the car ownership model in GMSPM is an adaptation of the current national model (MVA, 1996).

Table 3.3: Coefficients of DELTA (urban application)