Steerable Clustering for Visual Analysis of Ecosystems

Abstract

One of the great challenges in the geosciences is understanding ecological systems in order to predict changes and responses in space and time at scales from local to global. Ecologists are starting to recognize the value of analysis methods that go beyond statistics to include data mining, visual representations, and combinations of these in computational tools. However, the tools in use today rarely provide means to perform the kinds of rich multidimensional interaction that hold promise to greatly expand possibilities for effective visual exploration and analysis. As part of a project to develop a cyberCommons for collaborative ecological forecasting, we are developing ways to integrate highly interactive visual analysis techniques with data mining algorithms. We describe here our work in progress on steering mixed-dimensional KD-KMeans clustering using multiple coordinated views. Contributions include more flexible interactive control over clustering inputs and outputs, greater consistency of cluster membership during interaction, and higher performance by caching cluster results as a function of interactive state. We present our current tool that implements these improvements for visual analysis of Terrestrial ECOsystem (TECO) data collected from FLUXNET towers, with feedback on utility from our ecologist collaborators.