|
(21-09) Accounting for inter-laboratory variations in environmental standard setting using weighted hierarchical bootstrapping. Verdonck, Frederik*,1, Thas, Olivier1, Vanrolleghem, Peter1, 1 Ghent University, Coupure Links 653, Ghent, Belgium ABSTRACT- One of the problems with the current use of species sensitivity distributions (SSD's) is that information on inter-laboratory variations are discarded e.g. by averaging the data available for a particular species. In this poster, several alternative methods are proposed to account for this extra information. Depending on the interpretation of the inter-laboratory variations (i.e. variability or uncertainty), different weighted bootstrap sampling strategies were used. The first two methods consider inter-laboratory variations as variability i.e. inherent heterogeneity or diversity, which cannot be reduced through further measurements. The first strategy resamples from the entire pool of all weighted data points. The second strategy resamples different toxicity values per species. The third strategy considers inter-laboratory variations as uncertainty i.e. measurement error, which can partly be reduced through further measurements. It also resamples per species but only one sample for each species is taken. The three approaches result in similar SSD's but significantly different uncertainty bands. But most importantly, they have a different interpretation. In the first two cases, the SSD represents inter-species and inter-laboratory variability. In the last method, it only represents inter-species variability. In the first two cases, the uncertainty band represents sampling error. In the first case, the sampling error is due to sampling from a community whereas in the second case, the sampling error is due to sampling individuals from a species. In the last method, it represents sampling error and inter-laboratory uncertainty. Depending on the question raised, one or the other sampling strategy of the weighted hierarchical bootstrapping should be preferred. Overall, the first sampling strategy is preferred since it simulates the community diversity. Key words: species sensitivity distribution, uncertainty |
