To classify water bodies for stressor-response analysis, you identify groups of water bodies that have similar stressor-response relationships. Relationships estimated within the groups can be more accurate and precise than relationships estimated using the full data set. Classifying water bodies into distinct groups also provides the means to control for the effects of other variables (i.e., candidate classification variables) on the stressor-response relationship of interest while still allowing for relatively simple models that simulate the response as a function of stressor levels. For example, the coplot of hypoxic extent versus chlorophyll a concentration indicates that the stratification status of a lake strongly influences the slope of the relationships between chlorophyll a and hypoxic extent. This finding suggests that lakes should be classified by stratification status prior to stressor-response modeling. In this case, classification controls for the effect of the stratification and improves the precision of the resulting stressor-response relationships (see Figure 4).
Figure 4. Classifying lakes by the strength of stratification reveals a strong relationship between hypoxic extent and chlorophyll a in stratified lakes. The same relationship is much weaker in unstratified lakes.
Statistical approaches are available to automatically screen a large number of variables and identify a combination of classification variables that maximizes the precision of the stressor-response relationship of interest (e.g., TREED regression) (Yuan and Pollard 2014, Yuan and Pollard 2015a).
