Upscaling biodiversity: estimating the species–area relationship from small samples

The challenge of biodiversity upscaling, estimating the species richness of a large area from scattered local surveys within it, has attracted increasing interest in recent years, producing a wide range of competing approaches. Such methods, if successful, could have important applications to multi-scale biodiversity estimation and monitoring. Here we test 19 techniques using a high quality plant data set: the GB Countryside Survey 1999, detailed surveys of a stratified random sample of British landscapes. In addition to the full data set, a set of geographical and statistical subsets was created, allowing each method to be tested on multiple data sets with different characteristics. The predictions of the models were tested against the "true" species–area relationship for British plants, derived from contemporaneously surveyed national atlas data. This represents a far more ambitious test than is usually employed, requiring 5–10 orders of magnitude in upscaling. The methods differed greatly in their performance; while there are 2,326 focal plant taxa recorded in the focal region, up-scaled species richness estimates ranged from 62 to 11,593. Several models provided reasonably reliable results across the 16 test data sets: the Shen and He and the Ulrich and Ollik models provided the most robust estimates of total species richness, with the former generally providing estimates within 10% of the true value. The methods tested proved less accurate at estimating the shape of the species–area relationship (SAR) as a whole; the best single method was Hui's Occupancy Rank Curve approach, which erred on average by