Visual Parameter Selection for Spatial Blind Source Separation

Abstract

Multivariate measurements at irregularly-spaced points and their analysis are integral to many domains. For example, indicators of valuable minerals are measured for mine prospecting. Dimension reduction (DR) methods, like Principal Component Analysis, are indispensable tools for multivariate data analysis. Their applicability to spatial data is, however, limited. They do not account for Tobler's first law of geography, which states that "near things are more related than distant things." Spatial blind source separation (SBSS) is a data analysis and DR method developed specifically for spatial multivariate point data and hence accounts for such spatial dependence. SBSS requires analysts to set tuning parameters, which are themselves complex spatial objects: A partition of the spatial domain into regions and a point neighbourhood configuration. Their setting is dependent on each other, on domain knowledge of the analyst and data-driven considerations. We present a visual analytics prototype that guides and supports analysts in this process. We evaluated it with experts in visualization, SBSS, and geochemistry. Our evaluations show that our interactive prototype allows to define complex and realistic parameter settings efficiently, which was so far impractical. Settings identified by a non-expert led to remarkable and surprising insights for a domain expert.