You shall know a species by the company it keeps: leveraging co-occurrence data to improve ecological prediction

• DOI: 10.1101/2023.02.15.528518
• Publication Date: 2023-02-16T18:45:11Z
• Authors: Andrew Siefert; Daniel C. Laughlin; Francesco Maria S...
• Keywords: Trait; ENCODE; Global biodiversity

Abstract Making predictions about species, including how they respond to environmental change, is a central challenge for ecologists. Due the huge number of ecologists seek generalizations based on species’ traits and phylogenetic relationships, but predictive power trait-based models often low. Species co-occurrence patterns may contain additional information ecological attributes not captured by or phylogenies. We propose using ordination encode contained in species data low-dimensional vectors that can be used represent prediction. present an efficient method derive from GloVe, unsupervised learning algorithm originally designed language modeling. To demonstrate method, we GloVe generate nearly 40,000 plant statistics derived global vegetation dataset tested their ability predict elevational range shifts European montane species. Co-occurrence-based were weakly correlated with phylogeny, indicating unique Models co-occurrence-based explained twice as much variation only information. Given widespread availability occurrence data, learned are widely applicable powerful tool encoding many potential applications describing predicting ecology communities, ecosystems.