Accuracy of species richness estimators applied to fish in small and large temperate lowland rivers

摘要

The performance of non-parametric species richness estimators (Program SPADE) was assessed by applying them to fish from two lowland waterways in Poland: (1) a stream sampled annually at one site for 23 years (13 of which were after the stream was turned into a canal), and (2) a river sampled twice annually at two sites (one natural, the other impounded) for 16 years. On each sampling occasion consecutive electrofishing runs were made and the species richness of the total sample (obtained in all the runs) was predicted by each estimator from the sub-sample of the first run. The estimators were applied to all of the samples collected in each waterway, which were referred to as the 'rich group survey' selection, and to two smaller selections, named the 'improved survey' and 'complete survey'. The performance was evaluated using the measures PAR (percent of actual richness) and SRMSE (scaled root mean square error). Overall, the HM and Chao1-bc estimators were decisively better than others, and ACE1 and ACE were decisively worse both in terms of PAR and SRMSE. In the stream, the bed regulation little affected the performance of the estimators, but they were more correct when applied to the 'improved survey' selection rather than to the 'rich group survey' selection. In the river, the performance of most of the estimators, both in terms of PAR and SRMSE, was much improved only by selecting those samples for analysis that complied with the Chao-2 criterion (i.e., 'complete survey' selection).