Braverman, Amy (2003), Visual Data Mining for Quantized Spatial Data, Computing Science and Statistics, 35, I2003Proceedings/BravermanAmy/BravermanAmy.presentation.pdf
We discuss a visual data mining environment for quantized, multivariate spatial data sets generated by remote sensing instruments. Typically, remote sensing data are difficult to explore on a global scale because of their size, spatial, temporal and multivariate complexity, and hierarchical structure. We previously proposed (Braverman, JCGS March 2002) quantizing such data sets as a means of reducing their size and complexity. Specifically, data are partitioned on a spatial-temporal grid (e.g. one degree latitude by longitude by month), and data in each grid cell replaced by a set of differentially weighted representative values. The weights show how many of the original data points are represented by each representative. Here, we propose applying a modified version of the same methodology to the representatives themselves to achieve subsequent levels of data and complexity reduction. This allows us to coarsen both spatial and quantization resolution, facilitating better understanding of spatial, multivariate relationships at various levels of the hierarchy. We introduce a java visualization tool that allows interactive data exploration according to this model.