Quantitative relationships of invertebrates to pH in Norwegian river systems

The invertebrate fauna has been surveyed for twenty one unlimed generally acidic river systems in Norway. The data consist of 180 samples and 127 invertebrate taxa and associated water chemistry data (pH, calcium, acid neutralizing capacity, total aluminium, and conductivity). Multivariate numerical methods are used to quantify the relationships between aquatic invertebrates and water chemistry. Detrended canonical correspondence analysis (DCCA) shows one dominant axis of variation with high correlations for pH and aluminium. DCCA axis 2 is significantly correlated with calcium. The predictive abilities of invertebrates to pH are explored by means of weighted averaging (WA) regression and calibration and weighted averaging partial-least-squares regression (WA-PLS). The performance of the methods is reported in terms of the root mean square error of prediction (RMSEP) of (observed pH-inferred pH). Bootstrapping and leave-one-out jackknifing are used as cross-validation procedures. The predictive abilities of invertebrates are good (RMSEP_boot for WA = 0.309 pH units). Comparison of the invertebrates with diatom studies shows that invertebrates are as good predictors of modern pH as diatoms are. RMSEP_jack shows that WA-PLS improves the predictive abilities. Indicator taxa for pH are found by Gaussian regression. Anisoptera, Agrypnia obsoleta, Leptophlebia marginata, Sialis lutaria, and Zygoptera have significant sigmoidal curves where abundances increase with decreasing pH. Cyrnus flavidus shows a significant unimodal response and has an estimated optimum in the acid part of the gradient. Isoperla spp. and Ostracoda show significant sigmoidal responses where abundances increase with increasing pH. Amphinemura borealis, Diura nanseni, Isoperla grammatica, I. obscura, and Siphonoperla burmeisteri show significant unimodal responses and have high pH optima. Many taxa do not have statistically significant unimodal or sigmoidal curves, but are found by WA to be characteristic of either high pH or low pH. These results suggest that a combined use of Gaussian regression and direct gradient analysis is needed to get a full overview of potential indicator taxa.