Go to Paper
Return to GeoComputation 99 Index

Web-Based Multi-Agent Spatial Analysis Tools

MACGILL, James (J.Macgill@geog.leeds.ac.uk), OPENSHAW, Stan (stan@geog.leeds.ac.uk), TURTON, Ian (ian@geog.leeds.ac.uk), University of Leeds, Center for Computational Geography, Leeds LS2 9JT, U.K.

Key Words: smart spatial analysis, web, GAM, GEM, MAPEX, Flocks

The last few years have seen the development and enhancement of a number of tools for the analysis of spatial data. This has been a particular focus of the Center for Computational Geography of which GAM is the most well known. Other exploratory geographical analysis tools have been developed that use smart search methods while trying to track down patterns in GIS databases. Methods such as MAP Explorer (MAPEX) (which uses a genetic search procedure) and Flock (which uses swarm optimization) power fast and intelligent searches. This paper considers a different approach that is both novel and promises robust performance where multiple different pattern hunting exploratory search engines work together on a common problem.

The idea is simple. If there is a selection of tools to choose from for any given analysis application, then there is a much better chance of having one that will best suit the problem under analysis; however, the difficulty is knowing in advance which method will perform best in any given situation, as each different technology has its own advantages and disadvantages. In reality, it is likely that different techniques would not just perform differently on different data sets, but also on different parts of individual data sets and at different stages during the exploratory analysis. These problems are likely to become far more severe as the analysis task moves from purely spatial analysis to space-time analysis and higher dimensional data spaces. An ideal system would be able to take the best from each engine while overcoming the shortcomings of each method. This paper proposes a Multi-Engine Spatial Analysis Tool (MESAT) where the different search techniques can be made to work together in a single unified system in which the different search engines cooperate to produce an optimal hybrid spatial analysis technology. The conjecture is that the power of the whole will be greater than that of any of the individual components. The power comes from using multiple search procedures that cooperate by sharing their results.

The paper describes a three-tier architecture for MESAT, whereby the data to be analyzed is stored on a central server (acting as a data warehouse or a GIS), and is attached closely to this system as clients will be the different spatial search engines. Finally, there is a tier, which consists of a visualization system that allows users to observe and interact with both the results and the search mechanisms. This final tier is a web-based visualization system allowing all three tiers to be initiated, run, modified, and observed from almost anywhere in the world. This three-tier architecture allows multiple clients to tackle the same problem simultaneously, and then use multiple visualizations to look at the progress being made. The key to successful inter-search-client communication is a central shared result surface, which is built from the combined input of each separate search engine when they find interesting results. In addition, this surface acts as a form of message board allowing search engines to call in the help of other systems. For example, the fast but less rigorous systems could use the slow but exhaustive search engines when they encounter regions of the data set that look potentially interesting to ensure nothing is missed.

The paper also reports the results of applying this MESAT system to analyze synthetic highly complex spatio-temporal-multivariate data sets. The analysis of these sets would almost certainly have been out of reach if any one of the tools had been used individually.