Research Interests

My research interests can be grouped into three major themes:

• the development of algorithms for locational analysis and the investigation of new solution methods;
• the integration of spatial analysis and modelling with geographical information systems (GISs) to develop spatial decision support systems (SDSSs) for use by individuals and groups; and,
• the investigation and use of new computational environments (parallel and heterogeneous processing environments) to support the development of new algorithms and to improve human-computer interaction (HCI) in SDSSs.

Much of my work in these areas has formed part of research programmes at the U.S. National Center for Geographic Information and Analysis (NCGIA) and UCL's Centre for Advanced Spatial Analysis (CASA). Within the Department of Geography, I am a member of the GIS and RS Research Group and an associate of the Migration Research Unit. With the MRU, I have been working on how to integrate GIS and SDSS into migration information systems, especially based on the WWW (see EMIN, for example), to improve decision-makers' access to information and the tools to analyse it.

These three streams of work are mutually reinforcing because my research in locational analysis is the focus of my integration work with GISs, and the application of the resulting systems raises questions that require new algorithms to be developed. In both these streams of work, response times heavily influence users’ perceptions of HCI. Parallel processing and heterogeneous computing environments (HCEs) enable researchers to improve response times and to develop new ways of solving problems. Research into these computational environments supports the development of new locational algorithms, and architectures and implementation strategies for SDSSs.

The development of algorithms for locational analysis

Gerard Rushton (Iowa) and I developed a new data structure and implementation strategies that exploit the spatial structure of location-allocation problems. When applied to the vertex substitution algorithm and its derivatives, these strategies dramatically reduce solution times - making feasible the solution of very large problems. Building on this work, we designed a new algorithm, the Global-Regional Interchange Algorithm (GRIA), that produces solutions with the same characteristics the vertex substitution algorithm but does markedly less work. Solution times are reduced further and proportionally larger reductions in processing costs occur as problem size increases.

The data structures and implementation strategies in GRIA underlie the algorithms in the Locational Analysis Decision Support System (LADSS) - a prototype modelbase management system (MBMS) distributed by the NCGIA. I have investigated other types of heuristic solution methods, including genetic algorithms. Catherine Dibble (Ph.D. student, Buffalo) and I designed and implemented a genetic algorithm that solves location-allocation problems. Built around the same data structures as LADSS, this algorithm is effective but slow.

Spatial decision support systems

My early work on SDSSs focused on defining the characteristics of such systems and setting out system architectures. Subsequent work (much of it associated with the NCGIA’s Research Initiative co-led with Michael Goodchild) has addressed the problems associated with the design and implementation of the various components of SDSSs and improving the levels of human-computer interaction (HCI) that they support. A number of systems have been built to test solutions to these problems and applied in a range of problem domains.

Human-computer interaction

A SDSS must be able to display the results of analyses using appropriate maps and graphics. While most GISs produce thematic maps and graphics, they are limited in their support of domain-specific map types. Working with Marc Armstrong and others, I have developed a functional taxonomy of cartographic displays for use in locational decision-making to visualise different components of models and their results.
To explore their decision space, decision-makers must be able to combine analytical and graphical representations of their problem in flexible ways. Unfortunately, cartographic displays play only limited roles in the decision-making processes supported by many current SDSSs. HCI would be improved if systems provided multiple, simultaneous representations of a problem, users could interact with any of these representations, and the system responded by invoking the appropriate capability and automatically updated all available representations of the problem. The provision of such visual interactive modelling interfaces to SDSSs raises all sorts of system design issues and has been a major research area for me over the past few years. I have set out the basic requirements for visual interactive modelling in locational analysis and shown how the data structures in LADSS can be used to support multiple, linked representations. Because the underlying data structures are common, changes made to one representation can be used to update others. While the computational burden of supporting visual interactive modelling can be considerable, parallel processing and heterogeneous processing environments provide one solution to this problem. Maps can be constructed in parallel and algorithms within the MBMS can be implemented in such environments.

Collaborative Spatial Decision-Making

I co-led, with Marc Armstrong (Iowa) and Karen Kemp (NCGIA), the NCGIA Research Initiative Collaborative Spatial Decision-Making. This Initiative addressed the problem that while groups often are charged with addressing complex spatial problems, most GISs, SDSSs and spatial analysis tools are designed for individual use. My interests in this area focus on collaborative tools for modelling and visualisation.

More recently, I have been involved in setting a research agenda with the U.S. Interagency Group on Decision Support and the Aurora Partnership. This partnership has evolved into a coalition of researchers, practitioners, and decision-makers from federal, state, and local government agencies, educational institutions, private-sector entities, and NGOs aimed at facilitating the development and use of decision support tools, systems, and services for place-based management. This coalition, represented by PlaceMatters, seeks specifically to address the needs of policymakers, land and resource managers, and community leaders.

Applications

The LADSS MBMS has been used in SDSSs to reorganise services in both the U.S. and India. My contribution has been in terms of system design, the development and implementation of new algorithms, and field testing. Other applications I have worked on include a SDSS that integrates a suite of exposure-effects models with ARC/INFO. My contribution to this project was in system design and the development of the data management interface and scenario manager. I was involved in a two-year, UNDP-funded project to build a Public Distribution System: a GIS-based SDSS that incorporates location-allocation and routeing models and features a scenario manager. Two of the Indian development team came to UCL to work on the system’s design with me.

Within UCL’s Migration Research Unit (MRU), I have worked on a stream of research concerning how to provide access to statistical and other information sources for a diverse user base. I co-directed research on a feasibility study for the establishment of a European Migration Observatory and, subsequently, the European Migration Information Network, in which I took responsibility for the ICT-related aspects. My interest is in using GISs to visualise the information held in web-accessible databases so that users can locate and access information using maps and other visual methods and then visualise stocks, flows, and other types of information using a variety of representations, including maps and graphics.

I have worked with researchers at the Natural History Museum on the application of location-allocation models to the analysis of species distributions across Europe. Three papers on this work have been published: the first, in Ecography, initiated a dialogue in the literature that has continued in both Ecography and the Journal of Biogeography.

I have worked with Gerry Rushton and Joel Barkan, both University of Iowa, on the development of a new set of algorithms and associated SDSS capabilities to design electoral districts. We have published one paper on this work.

Parallel and heterogeneous processing environments

There are three motivations for examining parallel and heterogeneous computing environments: first, the computational characteristics of these processing environments mean that they are able to support new approaches to solving problems; second, with access to GISs, SDSSs, and disaggregate spatial databases, decision-makers increasingly wish to solve problems larger than traditional computing platforms can accommodate; and, third, new modelling environments, such as visual interactive modelling, greatly increase the computational demands of SDSSs.

A crucial step in the design and implementation of parallel software is to decompose a problem and its solution algorithm into parallel processes. Yuemin Ding (Ph.D. student, Buffalo), Marc Armstrong and I have worked to develop a theoretical basis for decomposition that exploits the peculiar structure of spatial problems; we have used this work to implement spatial algorithms with a range of computational characteristics: shortest path algorithms, a new hill-shading algorithm, and an algorithm that generates Delaunay Triangulations.

More recently, I have examined the use of heterogeneous processing environments to support SDSSs for locational analysis. Heterogeneous processing environments are suites of computers with different architectures, each of which has a set of processing characteristics that match the computational requirements of one or more parts of a complex problem. Marc Armstrong and I analysed the computing requirements of popular algorithms used to solve location-allocation problems and matched them with appropriate computer architectures. On the implementation side, we have looked at how to support both analysis and display with parallel processing; and I have integrated two forms of parallel processing computers with my LADSS software and developed a user interface that makes their use transparent to the user.