Classification of Reynolds phytoplankton functional groups using individual traits and machine learning techniques

The Reynolds Functional Groups (RFG) classification scheme is an informative and widely used method in ecological studies of freshwater phytoplankton. It clusters species with similar traits, as well as common environmental sensitivities and tolerances. However, researchers face the difficulty to classify species into RFG because it relies in expert opinion, taxonomical knowledge and environmental information, which are not always accessible. Thus, a step forward is to build general statistical models to classify species into RFG. Under the hypothesis that an organism's response to environmental conditionsdetermines their functional traits, here represented by the RFG, we predict that morphology and classification into broad taxonomic groups willexplain RFG independently from environmental information and expert knowledge. To evaluate the predictive ability of morphological traits (e.g. volume) and taxonomic affiliation (e.g. chroococcal Cyanobacteria) as discriminant variables of RFG, we compiled 1,300 species (264 waterbodies) and applied Random Forest (RF) and Classification and Regression Trees (CART). We divided the data to train the models and test their performance. RF successfully classified species into the 28 RFG (only c. 10% test error) with an average individual RFG success rate of 84.6 (range=33%-100%). This is a relatively high percentage of success from an ecological point of view. It suggests that the selected variables are able to reconstruct the RFG and represent well environmental preferences, without including information about local environmental conditions as classifiers. Our results reinforce the functional basis of the RFG and support both morphological traits and taxonomic classification as good proxies of phytoplankton responses to environmental conditions. A dichotomous key based on the CART was constructed, and an R code to classify species into the RFG is freely available. This work may help users to classify species into the RFG, including those that were not previously listed in the Reynolds classification system.