Predicting mayfly recovery in acid mine-impaired streams using logistic regression models of in-stream habitat and water chemistry.

Mayflies (Order Ephemeroptera) require high quality water and habitat in streams to thrive, so their appearance after restoration is an indicator of ecological recovery. To better understand the importance of restoring in-stream habitat versus water chemistry for macroinvertebrate communities, we developed taxon-specific models of occurrence for five mayfly genera (Caenis, Isonychia, Stenonema, Stenacron, and Baetis) inhabiting streams in the Appalachian Mountains, USA. Presence/absence records from past decades were used to develop single and multiple logistic predictive models based on catchment characteristics (drainage area, gradient), in-stream habitat variables (e.g., substrate, channel morphology, pool and riffle quality), and water chemistry. Model performance was evaluated using (a) classification rates and Hosmer-Lemeshow values for test sets of data withheld from the original model-building dataset and (b) a field comparison of predicted versus observed mayfly occurrences at 53 sites in acid mine drainage-impaired watersheds in 2012. The classification accuracies of final models for Caenis, Stenacron, and Baetis ranged from 50 to 75%. In-stream habitat features were not significant predictor variables for these three taxa, only water chemistry. Models for Isonychia and Stenonema had higher classification rates (81%) and included both habitat and chemical variables. However, actual occurrences of Isonychia and Stenonema at study sites in 2012 were low, consistent with the calculated probability of occurrence (Po ) < 0.60. Caenis occurred at test sites 35% of the time when the model predicted a Po  > 0.40. Stenacron showed the greatest consistency of actual versus predicted occurrences, occurring at 56% of sites when the Po (based on pH and conductivity) was > 0.50 and only at 1 site when Po  < 0.5. The results demonstrate how predictive models of individual indicator taxa could be valuable for evaluating the relative impacts of restoring physical habitat versus water chemistry during stream remediation.