Accuracy of non-parametric species richness estimators across taxa and regions

Abstract

AbstractNon-parametric species richness estimators are efficient and widely used when sampling is incomplete. There is little consensus on which of the available estimators works best across taxa and regions. Until now no work compared existing algorithms with multiple datasets encompassing contrasting scenarios.We used data from 62 inventories worldwide at different spatial scales, including 20 vertebrate, 22 invertebrate and 20 plant datasets, and compared the accuracy of the most used non-parametric estimators (Chao and Jackknife) and improvements to their original formulations.Our results highlight the good performance of the Jackknife estimators for incidence data, especially the P-corrected first order jackknife estimator (Jack1inP). This algorithm ranked most often the best or among the best performing estimators using two measures of accuracy that measure deviation from expectation along the accumulation curve.We argue that Jack1inP can be considered a universal estimator for species richness, regardless of taxon, temporal and spatial scales, or completeness of the sampling. More research should however be directed towards finding the precise contexts when each estimator might perform best.