Do we have to incorporate ecological interactions in the sensitivity assessment of ecosystems? An examination of a theoretical assumption underlying species sensitivity distribution models.

Species sensitivity distributions (SSDs) are statistical distributions which extrapolate single-species toxicity test results to ecosystem effects. This SSD approach assumes that ecological interactions between populations, such as grazing and competition, do not influence the sensitivity of ecosystems. The validity of this assumption in a simple freshwater pelagic ecosystem was tested using ecosystem modelling. For each of a 1000 hypothetical toxicants, a lognormal SSD was fitted to chronic single-species EC10s of the species present. As such, these distributions did not account for ecological interactions and were therefore termed 'conventional SSDs' (cSSDs). Next, sensitivity distributions that did take into account ecological interactions were constructed (eco-SSD) for the same 1000 toxicants, using an ecosystem model. For 254 of the 1000 hypothetical toxicants, mean and/or variance of the cSSD were significantly higher than mean and/or variance of the eco-SSD, as such rejecting the general validity of the tested assumption. A classification tree approach indicated that especially toxicants which directly affect phytoplankton (i.e. herbicides) may have a higher mean for cSSD than for eco-SSD. Conversely, means of eco-SSD and cSSD tend to be equal for toxicants directly affecting zooplankton and fish, e.g. insecticides. For the 254 hypothetical toxicants for which the tested assumption was false, a predicted no effect concentration (PNEC) calculated as the lowest single-species EC10 divided by an application factor of 10 was on average a factor 10 lower than the corresponding ecosystem-NOEC calculated by the ecosystem model.