79 
ANALYSIS ISSUES 
In conducting the analysis to formulate the decision rules and calibration curves that 
relate in situ turbidity measurements with calibration station K d measurements, 
numerous issues were addressed. Many of these issues focus on lumping or dividing 
the data when computing calibration curves. The argument in favor of lumping 
(performing the analysis on an aggregation of data) reasons that better estimates are 
obtained when averaging large numbers of observations. Lumping calibration data 
over time (e.g., one year) was assumed valid because the light-scattering properties 
of a tributary’s suspended sediments would remain relatively constant over time. 
On the other hand, the turbidity-to-K d relationship may prove inconsistent across 
different segments or entire tidal tributaries. After reviewing the Maryland and 
Virginia shallow-water monitoring data for 2003 to 2005, it was decided to divide 
the data into similar groups for individual calibration models and to conduct a cluster 
analysis for the group of tributaries monitored from 2003 to 2005. Algorithms for 
each group were developed that led to better overall precision. 
Other water quality parameters were tested for their ability to predict K d . Chloro¬ 
phyll and salinity from the calibration sites are also predictors of K d but their 
contribution is smaller than turbidity. Colored dissolved organic matter are likely to 
increase K d , however, these measurements are not routinely collected by the Chesa¬ 
peake Bay Water Quality Monitoring Program. Individual calibration curves may 
prove necessary for areas around the Bay where freshwater input from “blackwater” 
streams (e.g., the Pocomoke River) that drain extensive wetlands results in relatively 
high concentrations of colored dissolved organic matter. 
STATISTICAL MODELING 
The continuous turbidity measurements are calibrated to predicted light attenuation 
through the water column (K d ) by using statistical relationships among simultaneous 
measurements of turbidity, chlorophyll, salinity measurements, and light attenuation 
profiles of underwater photosynthetically available radiation (PAR) from five to 
eight calibration stations within each Chesapeake Bay segment. A multiple regres¬ 
sion model of K d vs. 1.5 root of turbidity [i.e., (turbidity) 1/15 ] x chlorophyll x 
salinity provides the best fit of the K d -to-turbidity relationship. The 1.5 root yielded 
the lowest root mean square prediction error and the highest r-square value. 
Figure VII-9 shows simple linear regressions of predicted K d versus the 1.5 root of 
measured turbidity for each of the seven Virginia tidal tributaries having shallow- 
water monitoring data from 2003 through 2005. Some of the slopes are similar but 
clearly different than others, indicating that data from small groups of tributaries 
with similar slopes can be combined into one calibration curve. 
The linear regression was further expanded to include terms for in situ chlorophyll 
and salinity. Like turbidity, the relationships between chlorophyll and K d , and 
salinity and K d vary among tributaries. However, enough similarities between 
chapter vii 
Shallow water Monitoring and Application for Criteria Assessment 
