Constructing species distribution models (SDMs) and ecological niche models (ENMs) for very rare species is a very hard challenge–and these are the species for which knowing their distributions and niches would help the most!
Recently, Kelley Erickson and I completed an assessment of state-of-the-art methods for modeling rare species in community contexts. In this analysis, we pit multi-species distribution models (hierarchical modeling of species communities, or HMSC, and species archetype models, or SAMs) against somewhat more traditional methods like ensembles of small models (ESM) and single-species GLMs. Multi-species models are supposed to be better in these situations because they allow the rare species to statistically “borrow strength” from more numerous species. We confronted these modeling methods with a virtual species with 2 to 64 occurrences, simulated within a community of real species.
We found that multi-species models were not always better. Rather, it really depended on the kind of task (modeling the niche or distribution), the position and breadth of the species’ niche, and of course sample size (see figure). In sum, borrowing of strength might actually be “borrowing of weakness”, if more common species are dissimilar from the rare species.
Modeling the rarest of the rare: A comparison between multi-species distribution models, ensembles of small models, and single-species models at extremely low sample sizes [open access]
Erickson KL and Smith AB. 2023. Ecography.