An appropriate ambition of the model-building component of urban science is the construction of the best possible comprehensive model which represents the interdependencies that make cities complex (and interesting) systems. To articulate this is to spell out a kind of research programme: how do we combine the best of what we know into such a general model? Most of the available ‘depth’ is in the application of particular submodels – notably transport and retail. If we seek to identify the ‘best’ – many subjective decisions here in a contested area – we define a large scale computing operation underpinned by a substantial information system that houses relevant data. Though a large task, this is feasible! How would we set about it?
The initial thinking through would be an iterative process. The first step would be to review all the submodels and in particular, their categorisation of their main variables – their system definitions. Almost certainly, these would not be consistent: each would have detail appropriate to that system. It would be necessary then – or would it? – to find a common set of definitions. It may be possible to work with different classifications for different submodels and then to integrate them in some way in connecting the submodels as part of a general model that, among other things, captures the main interdependencies. This is a research question in itself! It is at this point that it would be necessary to confront the question of exogenous and endogenous variables. We want to maximise the number of endogenous variables but to retain as exogenous those that can be determined externally, for example by a planning process.
There is then the question of scales and possible relations between scales. Suppose we can define our ‘city’, say as a city region, divided into zones, with an appropriate external zone system (including a ‘rest of the world’ zone to close the system). Then for the city in aggregate, we would normally have a demographic model and an economic model. These would provide controlling totals for the zonal models: the zonal populations would add up to those of the aggregate demographic model for example. There is also the complicated question of whether we would have two or more zone systems – say one with larger zones, one with finer-scale. Bit for simplicity at this stage, assume one zone system. We can then begin to review the submodels.
The classic transport model has four submodels: trip generation, distribution, modal split and assignment. As this implies, the model includes a multi-modal network representation. Trips from origin to destination by mode (and purpose) are loaded onto the network. This enables congestion to be accounted for and properly represented in generalised costs (with travel time as an element) – a level of detail which is not usually captured in the usual running of spatial interaction models.
A fine-grain retail model functions with a detailed categorisation of consumers and of store attractiveness and can predict flows into stores with reasonable accuracy. This model can be applied in principle to any consumer driven service particularly for example, flows into medical facilities, and especially general practice surgeries. This task is different if the flows are assigned by a central authority as to schools for instance.
The location of economic activity, and particularly employment, is more difficult. Totals by sector might be derived from an input-out model but the numbers of firms are too small to use statistical averaging techniques. What ought to be possible with models is to estimate the relative desirability of different locations for different sectors and then use this information to interpret the marginal location decisions of firms. This fits with the argument to follow below about the full application of urban science being historical!
In all cases of location of activities, it will be necessary in a model to incorporate constraints at the zonal scale, particularly in relation to land use. as these are applied, measures of ‘pressure’ e.g. on housing at particular locations can be calculated (and related to house prices). It is these measures of pressure that lie at the heart of dynamic modelling, and it is to this that we will turn shortly.
As this sketch indicates, it would be possible to construct a Lowry-like model which incorporated the best-practice level of detail from any of the submodels. Indeed, it is likely that within the hardy band of comprehensive modellers – Marcial Echenique, Michael Wegener, Roger Mackett, Mike Batty and David Simmonds for example – this will largely have been done, though my memory is that this is usually without a full transport model as a component. What has not been done, typically, is to make these models fully dynamic. Rather, in forecasting mode, they are run as a series of equilibrium positions, usually on the basis of changes that are exogenous to the model.
The next step – to build a Lowry-like model that is fully dynamic – has been attempted by Joel Dearden and myself and is reported in Chapter 4 of Explorations in urban and regional dynamics (which has recently been published by Routledge). However, it should be emphasised that this is a proof-of-concept exploration and does not contain the detail –e.g. on transport – that is being advocated above. It does tackle the difficult issues of moves: non-movers, job movers, house movers, house and job movers, which is an important level of detail in a dynamic model but very difficult to handle in practice. It also attempts to handle health and education explicitly in addition to conventional retail. As a nonlinear model, it does embrace the possibility of path dependence phase changes and these are illustrated (a) by the changes in initial conditions that would be necessary to revive a High Street and (b) in terms of gentrification in housing.
What can we learn from this sketch? First, it is possible to add much more detail than is customary but this is difficult in practice. I would conjecture this is because to do this effectively demands a substantial team and corresponding resources and, unlike particle physics, these kinds of resources are not available to urban science! Secondly, and rather startlingly, it can be argued that the major advance of this kind of science will lie in urban history! This is because in principle, all the data is available – even that which we have to declare exogenous from a modelling perspective. The exogenous variables can be fed into the model and the historians, geographers and economic historians can interpret their evolution. This would demand serious team work but would be the equivalent for urban science of the unravelling of DNA in biology or demonstrating the existence of the Higgs boson! Where are the resources – and the ambition – for this!!