Using museum specimens to estimate broad-scale species richness: Exploring the performance of individual-based and spatially explicit rarefaction

Estimates of spatial patterns of broad-scale species richness are central to major questions in ecology, evolution and conservation. Yet, they are scarce due to incomplete information on species distributions. Often the only germane data derives from museum specimens collected during non-standardized sampling. Rarefaction, a promising approach to estimate broad-scale richness with these data, estimates the expected number of species represented in subsets of n specimens drawn from N specimens collected in a sampling unit. One version of rarefaction, known as individual-based rarefaction, assumes that the N specimens collected in a sampling unit constitute a random sample of individuals in that sampling unit. Another version, known as spatially explicit rarefaction, assumes that the N specimens collected in a sampling unit are spatially aggregated. We examined the working hypothesis that, when applied to museum specimen data, spatially explicit rarefaction is less biased than individual-based rarefaction because it reduces overestimation due to spatially aggregated sampling. We derived five predictions from this working hypothesis and tested them using computer simulation experiments based on a database of 129,782 plant specimens from Nicaragua, and sampling units of 5 x 5, 50 x 50, and 100 x 100 km. One experiment was a negative control, whereby we simulated collection of randomly chosen individuals from each sampling unit. In contrast, three other experiments included spatially aggregated sampling. In all experiments we applied individual-based and spatially explicit rarefaction to estimate richness, with n = 200 and n = 500 specimens. As expected, the experiment designed as a negative control did not support the working hypothesis. The other three experiments supported the working hypothesis in analyses of larger sampling units, but not in 5 x 5 km sampling units. The predictions we derived from the working hypothesis can be used to assess which rarefaction version is best in particular systems.