Inferring the phosphorus levels of rivers from benthic diatoms using weighted averaging

Abstract

Diatom and water samples were collected from 145 river sampling stations in Finland. The relationship between benthic diatom taxa and measured environmental variables was explored using canonical correspondence analysis (CCA). A weighted averaging (WA) regression model was developed to infer phosphorus concentrations of river water. The optima and tolerances of the diatom species were obtained from the WA regression. The performance of the WA model was assessed using correlation coefficients and the root mean squared error of prediction (RMSEP) between the observed and inferred phosphorus concentrations. The model was cross-validated using an independent test set. The predictive power of the WA model was good (r = 0.91; RMSEP 13.9 μg P/l) in the training set. The weighted averaging also performed well in the independent test set (r = 0.87; RMSEP 15.6 μg P/l). The correlations were only slightly lower and predictive errors larger than in the training set. The weighted averaging gives a tool to evaluate trophic conditions in different ecoregions with the species optima list suitable for prevailing conditions and water quality.