TABLE 2:

Questions to consider when validating bioavailability models

Question	Comments
Is the validation study designed to test the accuracy of the model under the conditions for which the model was designed to perform?	A starting point for validation is to test the model with the species and conditions for which it was calibrated.
Are the organisms appropriate for testing the model?	Test organisms should be the species or species assemblages for which the model was parameterized/calibrated and to which the model is intended to be extrapolated.
Is the range of water chemistry appropriate for testing the model?	Bioavailability models need to be validated in a broad range of geographically and ecologically relevant water types, especially within waters that might produce high metal bioavailability. Consideration should be given to the range of chemistry conditions with which the model was developed.
Should the model be validated in a standard set of water chemistry conditions?	Models validated in a standard set of water chemistry conditions can provide a higher level of comparability among models; however, developing a standard test water for all metals will be especially challenging because factors that determine bioavailability can vary significantly among metals.
Is the model capable of predicting effects on the species or community in the water chemistry being tested?	Although bioavailability-based models are capable of predicting effects under laboratory conditions, less is known about performance under natural conditions in the field.
What level of extrapolation within or among taxonomic groups is appropriate?	Interspecific toxicity-prediction models can be used to predict responses among closely related species through the genus and even to the order level (e.g., US Environmental Protection Agency 2016b); however, extrapolation among major groups of organisms (e.g., cladocerans to aquatic insects) or across levels of biological organization is significantly more challenging because of variation in modes of action and mechanisms of toxicity.
Can the model be extrapolated across levels of biological organization?	Extrapolating across levels of biological organization is conceptually possible, and a few bioavailability models have been used to improve predictions of effects of metals on natural communities relative to hardness-based models in the field (Schmidt et al. 2010; Stockdale et al. 2010) and in mesocosm experiments (Iwasaki et al. 2013).
Can validation studies incorporate spatial and temporal variability occurring in the field?	Some modeling efforts that incorporate bioavailability and fluctuating concentrations of metals have been performed (e.g., Meyer et al. 2007a), but this needs more development.
Are multiple exposure routes for metals relevant to consider in validation of bioavailability models?	For some metals, physical effects (e.g., Al, Fe) or diet-borne exposure pathways (e.g., Hg, Se) might be necessary to consider in the validation study.
What is the accuracy goal for the model?	Validation needs to be rigorous but flexible enough to fit the user’s purpose. Support (or lack of support) for a model should be presented and reported in such a way that users can define their own level of acceptability (see Performance Evaluation section).
When is the model sufficiently validated?	Model validation is a continuous and iterative process that will necessarily be constrained by the amount of data available. Beck et al. (1997) provide guidance when deciding a sufficient level of model validation.