The classification schemes and modeling efforts will be endpoint-specific. For each endpoint in 
each of the regions and across all coastal systems, we will begin with a data mining effort and 
identify and collect data on key parameters and existing models and classification schemes that 
will provide the parts necessary to build a sound scientific basis for nutrient criteria for each of 
the three endpoints (DO, SAV loss, and food webs). With empirical data and knowledge of the 
key parameters that influence the nutrient-response curves, we can identify correlations and 
develop models to test hypotheses experimentally in the laboratory and in the field. This will be 
an iterative process to some extent until we have adequately characterized the endpoint/nutrient- 
response and controlling factor relationships needed to establish numerical nutrient criteria across 
the nation's coastal receiving waters. Once this is accomplished we can move on to watershed 
scales and/or other water body types. 
Complexity in ecosystem models is always dealt with by simplification; all models are 
abstractions of real systems. The simplifications range from complete linearization of the system 
with regression models to more complex relational models that incorporate a few critical 
measurements of the biological, chemical, and physical domain that makes up an estuary. Food 
web models make up the more complex non-linear models that include most of the mechanistic 
relations and feedbacks within the estuarine biogeochemical system. Relative to the coupled 
atmospheric-watershed-estuarine models developed for Chesapeake Bay and Long Island Sound, 
the food web models are easy to develop and have much fewer data requirements. 
Regression Models 
Regression models may be used to develop relationships among variables within a single system 
or among multiple systems. Such models use nutrient load to predict other parameters such as 
phytoplanktCMi biomass accumulation, primary production, sedimentation, and community 
metabolism, SAV loss, and DO-related response. This research will consist of cross-estuary 
analysis that focuses on common responses within classes of estuaries. Regression analysis that 
compares data among multiple systems has been used successfully for Maryland estuarine 
systems (Boynton et al. 1996); and a diverse collection of estuarine, continental-shelf, and open- 
ocean systems (Nixon et al. 1996). In the case of 37 side-embayments of Buzzards Bay, 
Massachusetts, such regression analysis has been applied to development of WQC and TMDLs 
for nitrogen (Costa et al. 1999). Such regressions are expected to have general application to 
systems similar to those for which they have been developed. These regressions will quantify 
estuary response to nutrient loading, and thus be directly useful in risk assessment and setting of 
nutrient criteria. In addition, regression models can be used as an adjunct to some of the 
proposed mesocosm and field work. For example, data from mesocosms to determine the effect 
of nutrient loading and benthic oxygen consumption on denitrification and nitrous-oxide 
production can provide insight into why some estuaries with nutrient sources having a high 
nitrogen/phosphorus (N/P) ratio remain nitrogen limited. Simple regression models also can be 
used in conservative mixing curves to determine sources and sinks of nutrients over the length of 
an estuary. Nitrogen, P, silicate, or other contaminants, when plotted against salinity, provide 
estimates of deposition, utilization, and supply of these materials over the length of an estuaiy. 
Vollenweider (1975) pioneered the development of regression models that allow extrapolation of 
data among systems using a scaling of biological processes to hydraulic residence time. This 
47 
