Estimating marine species richness: an evaluation of six extrapolative techniques

The number of species in an assemblage at a given point in time is a fundamental property of ecological systems, yet it is hard to quantify for many marine systems. We studied the performance of 6 techniques ('estimators') for extrapolating species richness from limited numbers of samples, using 3 datasets for which an absolute value for total species richness could be determined. We propose that the ideal estimator should always slightly overestimate species richness compared to any observed maximum species richness derived from sampling, as sampling error will always lead to underestimation of true richness. We quantified performance of the estimators relative to the sampled total richness in the assemblage across a range of efforts up to 80 % of that required to achieve the asymptote of the species accumulation. We used 3 measures: bias (mean deviation of an estimate from the known richness), precision (variance of repeated estimates based upon a subset of the available pool of samples), and overall accuracy (a combination of bias and precision). No single estimator performed best in all cases, and estimator performance was affected by sampling effort. The estimator Chao1 performed best at intermediate sampling efforts, with LAG S-infinity also performing well at high relative effort. S-infinity consistently underestimated, whilst Chao2 and ICE both overestimated and displayed poor precision and accuracy, especially at intermediate sampling efforts and in datasets with uneven patterns of species incidence. Species abundance and incidence amongst samples of a dataset were shown to affect performance of most of the estimators, with the exception of the recently proposed S-infinity family of techniques. We conclude that Chao1 represents the best compromise choice of estimator, and that such nonparametric techniques may represent useful tools for rapid estimation of species richness for some marine assemblages, based on limited sampling effort.