Biodiversity emerged as a paramount performance measure of organism traits or ecosystem functioning. Multidimensional functional indices, e.g. functional richness, functional evenness or functional divergence, are important quantitative summaries. These indices are evaluated on different community levels. Hence, the choice of the radii or scales determining community sizes is of predominant importance. Typically the radii are derived by an expert-driven procedure and/or mainly explained through the topography of the area. We alternatively propose to derive the radii in a completely data-driven manner, that permits being automatized and applicable to irregular lattice data. The method builds on ideas from image processing and identifies inherent and eminent spatial features. A Bayesian hierarchical model represents the field and is additionally exploited to handle missing values. The extension of the eminent features is derived through the precision matrix of the hierarchical model. Finally, the introduced approach is applied to the morphological traits of the mountain Laegeren in Switzerland and compared to the expert-driven indices radii.