Estimating probabilistic site‐specific species pools and dark diversity from co‐occurrence data

Abstract Aim The species pool specific for a site includes all species from the region that are theoretically able to live in the site's particular ecological conditions. The absent portion of the site‐specific species pool forms the site’s dark diversity, which is unobservable and can only be estimated. Most existing methods to designate dark diversity act in a binary fashion. Here, we argue that the species pool is more suitably defined as a fuzzy set, present a method to estimate probabilistic species pools using pairwise co‐occurrence data with a hypergeometric distribution, and compare it with established methods (Beals index and favourability correction). Innovation We compare the different aspects of the method’s performance using simulations based on individual agents in which the suitability for each species in each site is known. Further, we assessed the methods in two real datasets with nested sampling designs. We provide the R package ‘DarkDiv’ ( https://cran.r‐project.org/web/packages/DarkDiv/index.html ) that implements all the compared methods for estimations of probabilistic dark diversity and species pools. Main conclusions Beals method is extremely sensitive to species frequency, and predicts species’ suitability to local conditions less accurately than the other considered methods. The favourability transformation corrected this relationship, but still predicted extremely low probabilities for species with very little information. The hypergeometric method outperformed the Beals and favourability methods in all considered aspects in the simulations and displayed better characteristics in the real datasets. The hypergeometric method is currently the best option to estimate probabilistic dark diversity and species pool composition based on pairwise species co‐occurrence data.