To address this gap in knowledge, we tested the capacity of an environmental classification derived from the implementation of the European Water Framework Directive (WFD) to represent freshwater fish assemblages in a Mediterranean basin in South Western Iberian Peninsula. We determined the effect of using type-specific vs. site-specific approaches on the performance of different predictive models, as they are common tool for defining reference conditions. We compare predictive model performance using two different classification approaches (top-down vs. bottom-up) with the same predictive model (RIVPACS) and a continuum-like predictive model (ANNA).
Our results indicated that the top-down environmental classification did not group reference sites better than random, with high within-class heterogeneity and high between-class homogeneity. Although slightly better, the bottom-up approach did not show high classification performance. This result was not related to a weak biological response to environmental gradients; since biotic gradient was shown (i.e., there were gradual changes in fish assemblage along a longitudinal gradient). Consequently, the classification-based predictive methods failed to fit the biological data, and failed to predict the expected fish assemblages. Only the continuous predictive method succeeded and was accurate enough to be used to define reference condition.
Our results demonstrate that classifications that lack biological significance may have severe consequences for bioassessment programs. We recommend verifying the accuracy of classifications before using them as the basis for bioassessment under type-specific approaches. Alternatively, a continuous approach without pre-defined classes should be considered.