24 



Fritz (1990) used the canonical correspondence option of CANOCO 

 (ter Braak 1987) in a constrained ordination of 127 diatom taxa in 64 

 lakes of the northern Great Plains. The ordination axis was 

 constrained by the variable salinity, which resulted in a predictive 

 model that Fritz used in reconstructing historical changes in the 

 salinity of Devils Lake. 



In recent studies of Canadian lakes (Christie and Smol 1990, Hall 

 and Smol 1990), attempts to construct trophic predictive models 

 have involved canonical correspondence analysis as an explanatory 

 ordination method to identify limnological variables affecting diatom 

 assemblages in lakes having a wide range of trophic state but a 

 narrow range of pH. Weighted averaging calibration (Line and Birks 

 1990) was then used as the regression method to construct transfer 

 •^ functions and determine historical changes in trophic variables. 



, '-: Anderson (1990c) examined the weighted averaging approach to 



quantitative trophic-state reconstruction and warned that weighted 

 averaging studies largely utilize open water phytoplankton and v 

 chemical data, and that they ignore littoral community production 

 and chemistry. Anderson suggested that modeling methods should 

 be coupled with the use of multiple cores to calculate whole-basin 

 diatom accumulation rates that would give a reliable measure of both 

 plankton and periphytic paleoproduction. 



Comments on Statistical Methods 

 Cluster analysis is a statistical method that groups observations 

 into clusters that reflect their similarity without a priori 

 consideration of which factors are influencing the similarities. 



