Assessing similarity of n‐dimensional hypervolumes: Which metric to use?

Aim: The n‐dimensional hypervolume framework (Glob. Ecol. Biogeogr. [2014] 23:595– 609) implemented through the R package 'hypervolume' is being increasingly used in ecology and biogeography. This approach offers a reliable means for comparing the niche of two or more species, through the calculation of the intersection between hypervolumes in a multidimensional space, as well as different distance metrics (minimum and centroid distance) and niche similarity indexes based on volume ratios (Sørensen–Dice and Jaccard similarity). However, given that these metrics have con- ceptual differences, there is still no consensus on which one(s) should be routinely used in order to assess niche similarity. The aim of this study is to provide general guidance for constructing and comparing n‐dimensional hypervolumes. Location: Virtual study site. Taxon: Virtual species. Method: First, the literature was screened to verify the usage of the different metrics in studies (2014–2018) relying on this method. Subsequently, a comparative analysis based on simulated morphological and bioclimatic traits was performed, taking into consideration different analytical dimensions, sample sizes and algorithms for hyper- volume construction. Results: Literature survey revealed that there was no clear preference for one metric over the others in current studies relying on the n‐dimensional hypervolume method. In simulated data, a high correlation among similarity and distance metrics was found for all datatypes considered. For most analytical scenarios, using at least one overlap and one distance metric would be therefore the most appropriate approach for as- sessing niche overlap. Yet, when hypervolumes are fully disjunct, similarity metrics become uninformative and calculating the two distance metrics is recommended. The sample size and the choice of algorithm and dimensionality can lead to significant variations in the overlap of hypervolumes in the hyperspace, and therefore should be carefully considered. Main conclusions: Best practise for constructing n‐dimensional hypervolumes and assessing their similarity are drawn, representing a practical aid for scientists using the 'hypervolume' R package in their research. These recommendations apply to most datatypes and analytical scenarios. The R scripts published alongside this methodo- logical study can be modified for performing large‐scale analyses of species niches or automatically assessing pairwise similarity metrics among multiple hypervolume objects.