|
Key words: Geocomputational Methods, Visualisation, Integration, Scientific Process, Learning
GeoComputation encompasses a wide range of different tools and techniques, including data mining, knowledge construction, simulation and visualisation, all within the geographical realm. These activities take place along the entire extent of the scientific process, beginning with abductive tasks such as hypothesis formation and knowledge construction, through inductive tasks such as classification and learning from examples, and ending with deductive systems that build prescriptive models (that are common in spatial analysis and GIS). However, unlike GIS, there is no standard package or system that currently supplies these different types of functionality as an integrated whole; instead users must resort to a set of disparate (and often clumsy) programs that are difficult to connect together.
GeoVISTA Studio is a programming-free software development environment within which geocomputation and geographic visualisation tools are freely combinable. Studio employs a visual programming interface, allowing users to quickly assemble their own applications using a data-flow paradigm from a library of functionality implemented as Java Beans.
By combining visual and geocomputational approaches within the same environment, many benefits are realisable: Firstly, a visual interface allows abductive knowledge discovery agents to report their findings within a visual domain, thus drawing the expert's attention to potentially significant patterns within highly multivariate data spaces. Secondly, inductive learning agents can be trained in these visual data spaces from anomalies and structures recognised by human experts. Thirdly, visualisation allows the behaviour of machine learning tools to be monitored during training or configuration as a form of audit and control to ensure correct functioning (e.g., visualisation of hyperplane movement in neural networks).
At the time of writing, Studio contains full three-dimensional rendering capability and has the following functionality: Interactive parallel co-ordinate plots, visual classifier, sophisticated colour selection (including Munsell colour-space), spreadsheet, statistics package, self-organising map (SOM), and learning vector quantisation.
This paper discusses the roles that geocomputation and visualisation can play throughout the scientific cycle, emphasising their supportive and mutually beneficial relationship. A brief overview of the design of Studio is also given. Results are presented to show practical benefits of a combined visual and geocomputational approach to analysing and understanding complex geospatial data sets.