Evaluating Techniques for Interactive Exploration and Visualization of Large Astronomical Datasets

As large surveys of hundreds millions of objects are common, helping users locating their data subset of interest through interactive exploration and visualization is becoming a challenge of major concern. In this paper, we present two prototypes we developed to tackle this issue. Using Datavore and D3, we developed a pure-Javascript SPLOM (scatter plot matrix) visualizer taking a VOTable as an input. Linked views allow one to distinguish correlations between displayed attributes. This approach works well until 50k-100k objects, but does not scale beyond because of browsers limitations. For larger datasets, we adapted the Nanocubes datastructure initially created by AT&T Research for interactive visualization of spatial-temporal datasets. Our version, developed in Java, allows fast interactive visualization of a catalogue with hundred million rows for a few attributes. HiPS (Hierarchical Progressive Surveys) heatmaps are dynamically generated according to selected criteria and displayed in Aladin Lite. Eventually, we will discuss the benefits and limitations of these approaches, explore possible improvements and describe how these techniques might be integrated in existing CDS services.