A comparison of Nebraska reservoir classes estimated from watershed-based classification models and ecoregions

Regulatory agencies have been investigating a number of alternatives for classifying lakes into hydogeologically and ecologically similar assemblages that will facilitate establishment of attainable water quality standards. Concerns over the ability of traditional statistical classifiers to effectively classify environmental data have led to increasing interest in machine (predictive) learning classification tools such as decision trees. This paper compares the performance (classification strength) of a classification tree-based watershed classification model of Nebraska reservoirs to a discriminant analysis (DA)-based watershed classification system and reservoir classes derived from Omernik's Level IV Ecoregions. The performance of classification tree and DA-based watershed classification methods were also compared with respect to their cross-validation prediction errors. The results suggest that both watershed-based classification approaches (classification tree and DA) were more effective than Omernik's Level IV ecoregions in accounting for observed variations in water quality characteristics of Nebraska reservoirs. Moreover, this study demonstrates the utility of a classification tree algorithm, either as a supplement or alternative to DA, in handling the complexities of watershed variables and classifying Nebraska reservoirs for the purpose of water quality management. The classification tree also provides water resource managers with a useful interpretive classification interface.