Applying a GLM-based approach to model the influence of soil properties on the toxicity of phenmedipham to Folsomia candida

Soil properties are the main explanation to the different toxicities obtained in different soils due to their influence on chemical bioavailability and the test species performance itself. However, most prediction studies are centred on a few soil properties influencing bioavailability, while their direct effects on test species performance are usually neglected. In our study, we develop prediction models for the toxicity values obtained in a set of soils taking into account both the chemical concentration and their soil properties. The effects on the avoidance behaviour and on reproduction of the herbicide phenmedipham to the collembolan is assessed in 12 natural soils and the Organisation for Economic Co-operation and Development (OECD) artificial soil. The toxicity outcomes in different soils are compared and explanatory models are constructed by generalised linear models (GLMs) using phenmedipham concentrations and soil properties. At identical phenmedipham concentrations, the effects on reproduction and the avoidance response observed in OECD soil were similar to those observed in natural soils, while effects on survival were clearly lower in this soil. The organic matter and silt content explained differences in the avoidance behaviour in different soils; for reproduction, there was a more complex pattern involving several soil properties. Our results highlight the need for approaches taking into account all the soil properties as a whole, as a necessary step to improve the prediction of the toxicity of particular chemicals to any particular soil.