|
(21-27) Statistical and ecological evaluation of data sets for water quality criteria. Traas, Theo*,1, Mulder, Christian1, van Wijnen, Harm1, Aldenberg, Tom1, 1 RIVM PO Box 1 tp.traas@rivm.nl, Bilthoven, NL ABSTRACT- In The Netherlands, water quality criteria are preferably derived from the 5th percentile (HC5) of a probability distribution of screened toxicological test data. A previous study by Versteeg and colleagues showed that the HC5 for 11 compounds was conservative if related to the mean model ecosystem NOECs and seems to be a fairly good predictor of the lower 95% confidence interval. Recent statistical insights show that the HC5 uncertainty strongly depends on the number of toxicity data, especially for small data sets. Dutch quality criteria are sometimes based on small data sets. This finding prompted us to re-evaluate the statistical and the associated ecological uncertainty of the HC5. Other factors that influence the value of the HC5 are possible bimodality of the data set, and the method used to derive the 5th percentile (parametric or non-parametric). This study evaluates the uncertainty of the percentile that is chosen as a cut-off value for water quality criteria. The statistical uncertainty of the cut-off value is compared to model ecosystem endpoints for several compounds to estimate the associated ecological uncertainty. The goal is to determine which level of confidence of the HC5 is needed to guarantee sufficient protection. These procedures are demonstrated for several substances (agrochemicals, metals) for which quality criteria have been derived, and for which mesocosm data are available. In general, goodness-of-fit warrants the use of parametric methods, and non-parametric estimates are very close to parametric estimates. The confidence interval of the HC5 only occasionally overlaps with slight, transient mesocosm effects, and never contains serious ecological effects. The median HC5 is protective of (semi-)field effects in these case studies. Key words: water quality criteria, statistical evaluation, uncertainty, SSD |
