Abstract

Aim: Rarity—an important measure for conservation biogeography—can vary over many orders of magnitude. However, it is unclear which regional- scale abiotic condi- tions drive processes affecting rarity of endemic species on islands. To support con- servation efforts, we (1) assess the main abiotic drivers of endemic rarity, (2) determine how well existing protected areas (PAs) coincide with hotspots of endemic rarity and (3) introduce and evaluate a new hypervolume- based rarity estimator.

Location: La Palma (Canary Islands). Methods: We recorded all present endemic vascular plant species in 1,212 plots cov- ering the entire island. We calculated endemic rarity (corrected range- rarity richness for endemics) using a rarity estimation approach based on kernel density estimations (hypervolume approach). We performed a sensitivity analysis based on multiple linear regressions and relative importance estimations of environmental drivers to estimate the performance of the hypervolume- based rarity estimation compared to standard methods (occurrence frequency, convex hulls, alpha hulls).

Results: Climate variables (mean annual temperature, climatic rarity, precipitation vari- ability) best explained archipelago endemic (AE) and single- island endemic (SIE) rarity. Existing PAs covered the majority of AE and SIE rarity, especially national and natural parks as well as the Natura 2000 sites. In our study system, hypervolumes performed better than standard measures of range size.

Main conclusion: Both AE and SIE rarity on La Palma show a clear spatial pattern, with hotspots of endemic rarity found at high elevations and in rare climates, presumably owing to geographical and climatic constraints and possibly anthropogenic pressure (e.g., land use, introduced herbivores, fire). Areas of high rarity estimates coincide with the distribution and extent of PAs on La Palma, especially since the recent addition of the Natura 2000 sites. The hypervolume approach is a promising tool to estimate spe- cies range sizes, and can be applied on all scales where point/plot data are available.