Step 2, accurately classifying water bodies, applies only to the spatial reference condition approach. Group the set of reference water bodies identified in step 1 into classes by similar nutrient concentrations. Other factors such as elevation, geology, hydrology, positioning within the watershed, size and shape of the channel or basin also might be useful in defining your classes. Define each class to reduce the variability of nutrient concentrations within the class relative to variability among the classes. You can then specify different nutrient criteria that apply to waters for each individual class.

Classification can be an iterative process. You might have to revisit and refine your classification scheme as you progress through developing your criteria. For example, you might identify factors later in the process that affect nutrient dynamics and primary production responses.

• Using a geographic classification scheme (e.g., ecoregions, land resource areas, and physiographic provinces).
• Applying a statistical method biological, chemical, and physical data.

Geographic classification

EPA has developed a useful resource for initial classifications by delineating geographic divisions based on non-anthropogenic factors that affect ecosystems. You can use the maps the Agency has developed of ecoregions in the United States at various levels of resolution and aggregations based on interpretations of spatial coincidence of geographic phenomena that cause or reflect differences in ecosystem patterns (e.g., geology, hydrology, physiography, and soils). EPA has also developed aggregations of Level III ecoregions for use in nutrient management and assessment that reflect similar expectations for nutrient dynamics.

Statistical classification

You can also classify freshwater environments statistically by identifying characteristics of the reference water bodies associated with naturally expected nutrient concentrations. The first step in this process is to assemble data from reference waters that can predict naturally expected nutrient concentrations.

For lakes, physical classification factors might include water body morphologies (e.g., depths, widths, shapes, and volumes), water body origin, hydrodynamics (e.g., residence time), contributing watershed size, management regime (for reservoirs), and natural landscape and watershed vegetation. Chemical factors that might influence or be associated with naturally expected nutrient concentrations include alkalinity, color, hardness, iron, pH, silica, and temperature.

Stream classification factors might be based on canopy cover, channel morphology, flow continuity, retention time, slope, substrate features, and system size. Chemical factors similar to those for lakes also might be relevant for streams.

Physical classification of reference estuaries and coastal marine ecosystems can be more challenging because each water body is unique in terms of response to nutrient pollution, sensitivity, and size. For those water bodies, consider depth, habitat type (e.g., mangrove, seagrass, coral), morphometry, physical/hydrodynamic factors (e.g., circulation, mixing, stratification, water residence time, tidal amplitude, river flow, tides, wave exposure), and salinity gradients.

After assembling data on water body characteristics and nutrient concentrations in those water bodies, you can use multivariate statistical analyses such as cluster analysis or classification and regression tree to partition reference water bodies into groups with similar nutrient concentrations. Figure 1 shows the results of analysis in which TP concentrations at reference sites across the conterminous U.S. were modeled using a classification and regression tree with the following predictor variables: catchment area, elevation, longitude, latitude, mean annual temperature, and mean annual precipitation. The regression tree selected the predictor variables that minimized variance in TP concentrations within groups relative to among-group variance. Longitude was the first variable identified in the analysis, splitting the national data into western and eastern groups. Subsequent splits were found for latitude (in western streams) and for catchment area (in eastern streams).  Distributions of TP concentrations differ substantially among the groups, with streams in the northwestern U.S. having the lowest concentrations and large streams in the eastern U.S. having the highest concentration.

Figure 1. Results of a classification and regression tree analysis of reference site TP concentrations in the conterminous U.S. Boxplots below each group show the distribution of log transformed TP concentrations. Labels in the circles are as follows: Long = degrees longitude, Lat = degrees latitude, and Area = catchment area.

Examples of Classification Schemes

At this stage in developing nutrient criteria, you probably have compiled a variety of water quality, biological, and physical data used in previous efforts such as characterizing water bodies, selecting water body types, developing a conceptual model, defining reference conditions, and identifying reference water bodies. In this step, because the focus is exclusively on the reference water bodies you identified in Step 1, you will need to partition those data from the general population.

If you need additional data to effectively accomplish the classification task, you can access other information sources that include local, state, regional, and national databases. The Water Quality Portal EXIT is an example of a national database that is a clearinghouse for USGS, EPA, and USDA data.