97 



61% of the variance in areal macrophyte coverage. Another good 

 model for predicting percent-area coverage was the canonical 

 correspondence equation 3.8 that explained 45% of the variance in 

 this variable. I prefer the use of the canonical correspondence 

 analysis models over the multiple regression models because many 

 environmental factors may affect species' distributions (Patrick 

 1973), and interpretations based on a few species are less likely to 

 be reliable. None of these 4 models was statistically confounded by 

 covariant environmental or macrophyte variables. 



Macrophyte predictive models should prove useful for assessing 

 historical patterns in macrophyte standing crop that resulted from 

 changes in nutrient loading (Purohit and Singh 1985). Historical 

 water level changes might also be inferred from macrophyte 

 predictive models because macrophyte standing crop will increase 

 when lower water levels permit a lakeward expansion of rooted 

 macrophytes (Landers 1982), or when higher water levels permit 

 macrophyte colonization of a shallow littoral shelf (Anderson 1990). 



Floating-leaved biomass is best predicted using equation 3.5 that 

 explains approximately 87% of the variance in this variable using 5 

 diatom taxonomic groups. Partial correlations showed that the 

 floating-leaved biomass model was marginally unconfounded by 

 submerged biomass. None of the models for predicting submerged 

 biomass was useful, however, because of statistical confoundedness 

 with floating-leaved biomass. Perhaps this confoundedness occurs 

 because periphytic taxa are not specific in their attachment sites 

 with respect to floating versus submerged vegetation, and planktonic 

 taxa are negatively affected to an equal degree by both types of 



