The Temporally Integrated Monitoring of Ecosystems (TIME) Project Design: 1. Classification of Northeast Lakes Using a Combination of Geographic, Hydrogeochemical, and Multivariate Techniques

Abstract

This investigation is part of the Temporally Integrated Monitoring of Ecosystems ((TIME) project, an effort to meet the difficult challenge of monitoring surface water quality in the northeastern United States for signs of change in response to the Clean Air Act Amendments of 1990. The overall objective of the study was to develop a unified scheme for classifying lakes in the northeast into relatively homogeneous groups and improve the likelihood of detecting water quality trends in the region. The study approach involved combining the best elements of several procedures recently used for defining regional subpopulations of lakes; these were termed the hydrogeochemical model (HM), geographical model (GM), and multivariate statistical model (MSM). Lake and watershed data from the U.S. Environmental Protection Agency Eastern Lake Survey (ELS) were used to evaluate the classification methods and their modifications. After preliminary comparisons were made of the three classification schemes, it was concluded that the resulting subpopulations indicated that there was meaningful similarity among methods but that the significant dissimilarity reflected distinctive attributes of each classification method. These differences were deemed important; accordingly, integration of the methods entailed efforts to preserve parts of each. This was accomplished by assigning each lake of the ELS data set into a lake cluster that had been defined by jointly applying the HM and GM methods. Subsequently, the jointly classified clusters were aggregated by coupling an application of the MSM (cluster analysis) with subjective judgment regarding termination of the process of cluster formation. This integration of procedures gave rise to nine subpopulations that separated mainly on the basis of hydrogeochemical factors, though geographic influences also were evident in the results. The integrated classification procedure provided an explicit method involving the combination of several kinds of data to yield lake subpopulations. Although the process of integrating this information may stand alone as a useful exercise, the results obtained from the integrated classification model exhibited less dispersion than those formed by the parent procedures. This is an important consideration when subpopulation homogeneity is wanted as a means to improve trend detection. From the view of statistical efficiency, therefore, the integrated procedure may be considered to be more optimal than the parent schemes. When the complexity of the integrated approach is considered, however, we conclude that the next most precise classification scheme, the HM approach, may be preferred over the integrated classification for use in the design of an actual monitoring program.