Interactive Dendogram Visualization
Application to Analysis of Atmospheric Processes

I will briefly describe the process of acquiring data, initial data processing and several visualization techniques we have used. Preprocessing includes data filtering (separating data and noise) and further data reduction by k-means clustering. Our main present goal is to develop semi-automatic ways of classifying aerosol particles. Again, for this we use k-means clustering with varying distance metric. Visualization part, the one I am most responsible for, takes the result of previous analysis, builds hierarchy based on distances and presents the data via circular interactive dendogram Example operations the dendogram supports are: tree pruning based on size of clusters, time slicing, cluster search based on mass spectrum and of course, spatial and temporal zooming.

One can also automatically produce mass time frequency 3d plots of clusters and subclusters and visually scan individual particles of a cluster (this is important for the evaluation of distance metrics that produce these clusters). Some current and future work is to incorporate an expert input into the system, to add some kind of non-linear zooming (fish-eye zooming is an example of this), automatic distance metric generation based on expert input, maybe experimenting with different ways of classification like support vector machines, etc.



PowerPoint slides relevant to this presentation

Top of Page