Can we reliably estimate species richness with large plots? an assessment through calibration training

The number of species (species richness) is certainly the most widely used descriptor of plant diversity. However, estimating richness is a difficult task because plant censuses are prone to overlooking and identification errors that may lead to spurious interpretations. We used calibration data from the French ICP-level II plots (RENECOFOR) to assess the magnitude of the two kinds of errors in large forest plots. Eleven teams of professional botanists recorded all plants on the same eight 100-mA(2) plots in 2004 (four plots, eights teams) and 2005 (four plots, nine teams including six from 2004), first independently and then consensually. On average, 15.5% of the shrubs and trees above 2 m were overlooked and 2.3% not identified at the species level or misidentified. On average, 19.2% of the plant species below 2 m in height were overlooked and 5.3% were misidentified and 1.3% were misidentified at the genus level (especially bryophytes). The overlooking rate also varied with plant species, morphological type, plot and team. It was higher when only one botanist made the census. It rapidly decreased with species cover and increased with plot species richness, the recording time of the census in the tree layer and the number of the censuses carried out during the day in the ground layer. Familiarity of the team with the local flora reduced the risk of overlooking and identification errors, whereas training had little impact. Differences in species richness (over space or time) in large plots should be cautiously interpreted, especially when several botanists participate in the survey. In particular, the quality of the data needs to be evaluated using calibration training and, if necessary, may be improved by involving more experienced botanists working in teams and by fixing a minimum recording time.