Exploring patterns of global vascular plants occurrence data with nonlinear dimensionality reduction

Multiple historical and contemporary processes drive species distribution. Understanding the role of these drivers is an important task in the context of ongoing environmental change. However, data on biodiversity are often represented by taxonomically or geographically limited datasets. These drawbacks prevent us from testing the relationships between biodiversity distribution patterns and driving factors across a variety of environmental conditions.  In this context, species distribution data of broad spatial and taxonomic coverage published via Global Biodiversity Information Facility (GBIF) provide new possibilities, which have not been extensively explored. Here we aim to derive plant co-occurrence gradients from the global GBIF dataset using nonlinear dimensionality reduction methods.

We analyze global vascular plant co-occurrences using the Isometric Feature Mapping (Isomap) method, which requires no a priori assumptions and allows effective representation of high-dimensional datasets. The study seeks answers to the  following questions: (1) to which extent GBIF records can be considered representative in comparison to curated floristic datasets, i.e. checklists; (2) how the biases of GBIF data should be addressed for a large-scale analysis; (3) how the co-occurence gradients from binary presence-absence data are compared with those that derived from relative abundances. The study  highlights the potential of constantly-growing open access biodiversity data for large-scale analysis and cross-validation. We also anticipate the results to provide support for advanced interpretation of macroecological patterns from mostly opportunistic plant observations.