Estimating regional species richness using a limited number of survey units

The accurate and precise estimation of species richness at large spatial scales using a limited number of survey units is of great significance for ecology and biodiversity conservation. We used the distribution data of native fish and resident breeding bird species compiled for two geographic regions in the U.S.A. to evaluate five established (Jackknife-1 and -2, Chao-2, ICE, and Bootstrap methods) and two new (CY-1 and -2) estimators. Both new estimators are based on relationships between species richness per subsample and the mean Jaccard coefficient across multiple pairs of subsamples, but they differ in the way the relationships are fit. The four regional faunas (two regions x two taxonomic groups) exhibited distinct species-occurrence distributions and a range of spatial heterogeneity. Re-sampling techniques were used to generate subsamples of five sizes (0.61-11.5% of a whole region) for examining the effect of sampling effort. With the total number of species recorded in each region taken as the regional richness, CY-1 and -2 were least biased at low sampling effort and CY-2 and Jackknife-2 were least biased at higher sampling effort. The differences in performance could be partially attributed to whether an estimator relied on the number (e.g., Jackknife-1) or the proportion of singletons (CY-1 and -2) for extrapolation. The estimation of fish species richness was more biased and less precise than that of bird species richness. This difference was closely related to how species-occurrence probability varied among species in a fauna (i.e., species-occurrence probability distribution). The estimators tested, particularly CY-2 and Jackknife-2, are useful in estimating regional total species richness; however, more robust methods are needed, which should take the form of species-occurrence probability distributions into account.