Modelling outperforms typologies for establishing reference conditions of boreal lake and stream invertebrate assemblages

Continuous modelling and discrete classification (aka typology) are two approaches commonly used to partition natural, spatial variability, and ultimately gauge anthropogenic effects on biodiversity loss and other valued ecosystem services. Using benthic invertebrate assemblages of boreal lakes and streams, we tested the efficacy of continuous modelling and discrete classification for partitioning natural variability of sites judged to be in reference condition. We anticipated that species distributions and assemblage composition would be more accurately predicted by models in general and specifically that models based on suites of predictor variables would outperform models based on a limited number of variables. Furthermore, we predicted that more flexible typologies would perform better than approaches using sets of mandatory categorical variables. Our results showed that models were more accurate at estimating species distributions and assemblage composition than typologies. Furthermore, models calibrated with only a few typology based variables were as accurate as full models, indicating that the main environmental gradients were captured by the classification variables used in our study. Continuous modelling also had lower incidences of false positives (<7%) compared to typological approaches (3.8-56%), i.e. a lower frequency of classifying reference sites as possibly impaired. The findings that continuous modelling outperformed discrete classification and that the latter had substantially higher frequencies of false positives is somewhat disconcerting given the relatively widespread use of typologies in bioassessment and management. Misclassification results in the unnecessary use of resources to re-classify sites, or more seriously implementation of unwarranted measures of rehabilitation.