52 
I don’t believe there is sufficient gain from adding month to warrant the degree of 
splitting of the data that will be required by doing monthly calibration curves. 
6. Can Tributaries be grouped so that calibration terms are uniform within group? 
At this point, we have established that the model should include three useful predic¬ 
tors: turbidity, chlorophyll, and salinity. These are terms suggested by Chuck 
Gallegos of the Smithsonian Environmental Research Center, Edgewater, MD, 
(personal communication) as likely to be important. The question now is whether or 
not there are groups of tributaries where the intercept and the coefficients for these 
three predictors are fairly uniform so that they may be lumped for one calibration 
model. The coefficients are shown in Table D-6. Clearly trying to organize these into 
uniform groups is complex. To assist with this organization, a cluster analysis was 
implemented where the tributaries are the items clustered and the coefficients are the 
attributes to cluster by. Note that because some coefficients are large, but not statisti¬ 
cally significant. These data were filtered by statistical significance before clustering 
by setting all coefficients with p-value >0.1 to zero. Note for example the salinity 
coefficient for the Potomac. At 4.3, the coefficient is nearly two orders of magnitude 
greater than other salinity coefficient and yet it is not even close to being statistically 
significant (p=0.74). The sample size for the Potomac is fairly small and the salinity 
range for the data collected is also small. These factors contribute to this aberrant 
coefficient. This illustrates a hazard of splitting data into subsets that are too small. 
The results of the cluster analysis are illustrated by the dendrogram in Figure D-l. 
Figure D-1. Dendogram illustrating clustering of Maryland Tributaries by model 
coefficients. 
appendix d • Derivation of Regressions 
