Diatom-based environmental inferences and model comparisons from 494 northeastern North American Lakes

The relationships between diatom assemblages and important limnological variables were investigated in 494 lakes from northeastern North America (Pennsylvania, USA, to Nova Scotia, Canada). The limnological variable most significantly related to diatom assemblages was lake water pH, although dissolved organic carbon and nutrients were also important. Based on these strong relationships, highly significant diatom-based inference models were developed to reconstruct key limnological variables based on diatom assemblages using weighted averaging (WA), maximum likelihood (ML), and modern analogs technique (NUT). The performances of the pH-inference models were high, similar, and significant (WA: r(boot)(2) = 0.89, root mean squared error of prediction (RMSEP) = 0.43; ML: r(boot)(2) = 0.89, RMSEP = 0.45; MAT: r(boot)(2) = 0.89, RMSEP = 0.46). In addition, distribution of sites along a pH gradient did not have the anticipated bias, especially with respect to the WA model, although an evenly distributed study set did result in slightly less noise. While some regionally specific information may be lost by utilizing a large number of lakes from a wide geographic area, the broad limnological gradients allow a more realistic, accurate, and complete description of the ecological characteristics of diatom species. In addition to providing new autecological data on diatoms for northeastern North America, these inference models can now be used to infer accurately and precisely lake water pH and associated variables for limnologists and lake managers.