BEST SUBSET MODELING OF PHOSPHORUS IN THE GRAND-RIVER USING CORRELATED VARIABLES

Statistical models are developed for the estimation of phosphorus conc entrations for different seasons and locations using routinely monitor ed water quality parameters in the Grand River Basin. Two methods of m odelling, namely the best subset model and the stepwise regression met hod based on R(2) and F values, are described. The best subset modelli ng procedure enables comparison between full model (containing all the independent variables) and subset models (containing subsets of indep endent variables). For correlated independent variables, the best subs et modelling procedure is shown to provide a better model than the ste pwise regression procedure. The statistical modelling results indicate that suspended solids play an important role in the prediction of pho sphorus levels and consequently decreasing suspended solids would decr ease the growth of aquatic plants in the Grand River Basin.