Assessing species misidentification rates through quality assurance of vegetation monitoring

Most studies of observer discrepancies in vegetation recording have been limited in the extent to which they can separate different sources of error. It is straightforward to quantify the degree of disparity between two species lists but not clear how to allocate a particular discrepancy to a specific cause. Misidentification is especially difficult to detect, and is rarely discussed in the literature. The vegetation monitoring protocol devised by the United Kingdom Environmental Change Network (ECN) splits each plot to be recorded into cells, within each of which a species list is compiled. This provides an objective measure of the frequency of occurrence of individual species, in place of the more subjective estimation of cover, and allows within-plot variation to be quantified. An added advantage of the ECN methodology is that botanical expertise and the use of cells can be combined in quality assurance (QA) studies to detect instances of consistent misidentification of species, thereby increasing the repeatability of vegetation recording and enhancing the possibility of detecting change. This paper reports an analysis of the data obtained from a 1996 ECN QA exercise and describes the methods used to pinpoint the most likely sources of discrepancies between the original site surveys and the QA survey. Overall it is estimated that 5.9% of specimens were misidentified at species level and 1.9% at genus level, though it is detectable that sites employing consultant surveyors achieved slightly better results. Misidentification rates are particularly high for the lower plants and for woodland plots. The number of unmatched records (the pseudoturnover rate) is high, 24%, but comparable to other studies. This does not seem to be the result, to any great extent, of seasonal changes or identification problems, but appears to be largely due to overlooking and partly a result of relocation problems. The overall percentage agreement between surveyors was 57%, also comparable with other studies.