Functional unsupervised classification of spatial biodiversity

Abstract

Abstract The recognition of spatial heterogeneity as well as of areas of low and high biodiversity through spatial techniques is essential to guide decision-making regarding the conservation and management of natural areas. In this context, reliable maps of biodiversity across sampling sites can be useful tools. Many ecological studies, which have dealt with a spatial approach for biodiversity, have focused only on one specific biodiversity aspect at a time, such as species richness or species evenness, yielding a partial overview of this complex concept. To solve this issue, we propose a spatial functional data analysis approach to diversity profiles for assessing spatial biodiversity and identifying groups of sampling sites which are similar in spatial patterns. Specifically, the functional distance-based LISA algorithm has been extended to the case of diversity profiles in lattice, after smoothing the discretized curves and specifying a suitable distance measure. The proposed spatial clustering algorithm has been applied to a real data set involving tree species diversity in a fully censured plot in the Harvard Forest, New England region. Our approach provides a useful method for identifying areas of low and high biodiversity, with the potential to address the monitoring of environmental policies. Indeed, we think that a classification of diversity profiles, which takes into account the spatial dependence, would permit a more homogeneous partition of sampling stations with a substantial noise reduction in supporting conservation planning.