SCALED CHRYSOPHYTES (CHRYSOPHYCEAE AND SYNUROPHYCEAE) FROM ADIRONDACK DRAINAGE LAKES AND THEIR RELATIONSHIP TO ENVIRONMENTAL VARIABLES1

The relationships between 23 scaled chrysophyte taxa (Chrysophyceae and Synurophyceae) and measured limnological variables in 62 Adirondack, New York, drainage lakes were examined by canonical correspondence analysis (CCA). The major proportion of variation in chrysophyte species distributions was strongly related to total monomeric Al (Al_m) and Mg concentrations, and their close correlates pH, Na, Ca, and acid-neutralizing capacity (ANC). Total monomeric Al concentrations explain a greater proportion of species variation than pH, suggesting that Al_m concentrations may be more important in governing the distribution of chrysophyte taxa in these lakes. Gaussian logit (GL) and linear logit (LL) regressions of the relative percentages of individual chrysophyte taxa to lakewater pH and Al_m concentrations and the examination of pH–Al_m response surfaces show that many chrysophyte taxa exhibit unique responses to these environmental gradients; taxa can be characterized as alkaline, circumneutral, acidic, and pH indifferent. Within each of these groups, taxa can be characterized further based upon their optima and tolerances to Al_m concentrations. Chrysophyte indicator species (i.e. a taxon with a strong statistical relationship to the environmental variable of interest, a well-defined optimum, and a narrow tolerance to the variable of interest) for pH include Mallomonas hindonii, M. crassisquama, M. pseudocoronata, and Synura uvella; M. hindonii, M. crassisquama, M. pseudocoronata, S. petersenii, and S. spinosa are good indicators of Al_m concentrations. Highly significant predictive models were developed to infer lakewater pH and Al_m concentrations from the relative percentages of chrysophyte scales in the study lakes. Model evaluation was based on their correlation coefficients and the root-mean-squared error of prediction (RMSE) derived from bootstrapping. Weighted averaging regression and calibration with tolerance down-weighting (i.e. weighting taxa inversely to their variance) produced superior results when compared to the computationally and data-demanding maximum likelihood methods and to simple weighted averaging regression and calibration.