17 



transformed index B values with pH for a set of 30 Swedish lakes 

 and produced a model with which they assessed lake acidification 

 due to atmospheric deposition in Sweden. Index B is somewhat 

 statistically dubious, however, because coefficients for the 

 autecological terms could not have been calculated by a simple linear 

 regression between pH and log index B as indicated by Renberg and 

 Hellberg (Whitmore 1989). 



Cluster analysis has been used to identify diatom assemblages 

 characteristic of various pH conditions (e.g. Davis and Anderson 

 1985). Charles (1985) used cluster analysis to group diatom species 

 with similar pH requirements and he performed a multiple 

 regression of these clusters with pH values of 38 Adirondack lakes. 

 His model explained approximately 90% of the variance in pH in his 

 calibration data set. 



Davis and Berge (1980) performed a stepwise multiple 

 regression of 33 taxa in a set of Norwegian lakes and produced a 

 model consisting of 7 taxa that explained 93% of the variance in pH 

 (unadjusted R2). Dixit and Evans (1986) have shown, however, that 

 particularly in lakes with spatial variability in diatom assemblages, 

 the use of indicator assemblages rather than individual taxa in 

 predictive models will greatly reduce the error term. Hustedt's 

 (1937-38) pH autecological categories have also been used in 

 multiple regression equations to develop pH predictive models based 

 on diatom assemblages (Davis and Berge 1980, Charles 1984, 1985). 



Ordination techniques, which reduce the number of diatom 

 variables in a model, have been used to construct pH predictive 

 equations. Principal components analysis is an indirect ordination 



