The Use of Visualization for Modelling and Simulation — Michael Batty

To set the context for exploring visualization for social science modelling, this paper examined first different types of models identifying iconic/pictorial, data-driven, and mathematical/simulation as those of significance to visualization. In fact, sometimes icons such as 3D digital structures of cities, and data-driven systems based on 2D additions of data are often called models but without denigrating the place of these, this paper concentrated upon mathematical/simulation models which we felt were of more concern to the social sciences per se. Such models can be classified according to whether or not they have a natural spatial-visual structure, this biasing visualization potential to geographic-built form-natural environment views of social systems. Natural spatial- visual models, non-spatial-visual models with some spatial components from which visualization can be exploited, visual abstractions from non-spatial models, and visual abstractions from spatial models were the four types identified. To this questions of temporal and model process dynamics which imply related types of visualization were added. Visualization itself was discussed in terms of dimensionality of the model, the extent to which abstract/real 'maps' might be constructed, abstract/real charts/graphs might be constructed and the extent to which new pictorial media might be added. Linking visualizations and data through hotlinking was discussed and then visualization in models was talked of in terms of inputs, outputs, model processes and mechanics and finally user interfaces for model-building and for communication. These ideas were to frame the subsequent discussion in the meeting.

This paper discussed two urban visualisation projects undertaken at Cardiff. The first, raises issues relating to computing requirements for computationally intensive 2D visualisation experiments. The principal aim of this project, now completed, was to develop methods for measuring detailed micro-urban form using pattern-recognition techniques; an aim originally motivated by the quest for automated feature recognition and learning in intelligent urban GIS but also relevant to the formal measurement of urban morphology. The project entailed coding a vast suite of pattern recognition algorithms within a framework suitable for conducting experiments on urban satellite imagery and co-referenced GIS layers. At the time no commercial package offered quite the range of approach and flexibility of measurement design required and the programmes were coded in Pascal. Visualisation was achieved with UNIRAS and the two programmes were loosely coupled via ASCII file exchange. The main computational problems encountered were the huge CPU time and space requirements which, at the time, exhausted the capacity of Cardiff's most powerful computer; the inefficiency of loose coupling between analytical and visualisation environments; and UNIRAS' inflexibility as a graphics package. Wider issues arising from this experience and discussed during the workshop include:

(a) the usefulness of 2D visualisation in designing urban morphological measures, demonstrated for example, by the creation of UNIRAS images of the vector outputs of NAG's 2D fourier transform routines in order to explore urban morphological regularities in Fourier space;

(b) the inevitability of having to put together bespoke programming environments to support this kind of work and the need to recognise this in research council grant applications;

(c) the need for super computing facilities in the social sciences; and

(d) the need for resources to ensure that the results of extensive programming effort are captured for the wider community.

The second project presented develops the theoretical and methodological specification for experiments that use a cellular automative model to explore the impact of alternative systems of pollution property rights on urban form and structure. Cities are grown in cellular space under the assumptions of profit maximising developers and welfare maximising households. Profit functions and externality functions are specified and both made a dynamic function of local neighbourhood conditions and sub-regional conditions in the cellular space. Equilibrium conditions are specified that relate to alternative regulative regimes. These generate profit maps which guide the simulation under Monte Carlo rules. The use of cellular automata allows investigation of the sequencing and spatial dimensions of the social efficiency questions embodied in the economic models. This presentation illustrated just how vital visualisation is to generative modelling of this kind in which interesting global patterns arise from probabilistic models of the local behaviour of elementary actors. In the experiment reported, clear spatial patterns (linear industrial zones) emerge from an essentially aspatial economic model of externalitiy internalisation. The patterns that form the outcome of the simulations can only be understood by visualisation.

The aim of this paper is to show that visualisation, as a dimension of 'configurational analysis', can be used not just as an aid to communication or understanding but to discover and demonstrate objective 'deep structures' in real phenomena. The field of phenomena are architectural and urban systems, alias the built environment, seen as organised forms and spaces. The idea of configurational analysis is first explained, then a simple example is shown of how cultural patterns in domestic space can be detected and visualised, followed by a complex example showing how deep spatial structures in urban systems can be detected and explained by a process involving visual representation.

The workshop offered the opportunity to present ideas to other social sciences in different subject areas. This is useful to develop thoughts and exchange information

There is a need for advocacy to the ESRC to encourage visualization research in modelling — priority of domains concerning cities, human geography, built environment. This needs to be set in the context of priorities for other social science research.

Working across the WWW will have an impact on social science research and modelling. We need to investigate the technologies, e.g. Java and make people aware of the research applications possible. Collaborative research, sharing of data etc are all of potential benefit.

There is an emergence of microbased data. Our understanding is changed if we have finer grained data. Models can provide greater understanding of patterns and processes with use of suitable techniques and data. There is a large underworked area in looking at networks (e.g. social networks) using visualization techniques.

What are the possibilities for using visualization in qualitative data analysis — perhaps video clips in ethnographic studies?

ESRC is encouraged to recognise the importance of visualization research in modelling.

There is a need to raise awareness amongst researcher of the potential of visualization software tools.

There is a need for a study to research the current use of visualization technology worldwide.

We need to illustrate potential use of tools using case studies. A workbook illustrating generic graphics for social scientists would be useful.