| I developed and evaluated methodologies for studying ecological communities. The first component of my work, Chapter 2, develops a scientific framework, referred to as the nutshell philosophy, for classifying aquatic habitats in the St. Croix National Scenic Riverway. The "nutshell" is a compact philosophy for practitioners, including two parts: (1) a classification framework using a small set of environmental predictor variables; and (2) test of the classification using biological data. My classification resulted in two units: segments and reaches. Segments have a dimension of 15+ km and are delineated by major tributary outlets and channel slopes. Reaches are nested within the segments , with a dimension of 2+ km and are delineated by changes in stream substrate. A data set of mussel communities was clustered to test the classification. The clusters were consistent with the segments. The "nutshell" philosophy was validated in this case.;The second component of my work, Chapter 3, introduces a method investigating stream substrate with underwater videos. With it a desired large area can be covered, and the resulting videos can be archived and used for quantitative analysis.;The third component of my dissertation, Chapters 4 and 5, relate to clustering methods. Chapter 4 investigates impacts of selected cluster number on the accuracy of clustering algorithms. I clustered data sets of known structure with three different methods, varying the cluster number; clustering accuracy was evaluated using the Rand statistic. The Elbow phenomenon was typically found for the response of the Rand statistic with the cluster number. I described the Rand curves with an analytical relationship and proposed a threshold slope of 0.001 to locate the optimal cluster number.;Chapter 5 evaluates performance of an emerging clustering method, Self-Organizing Map (SOM), with the more traditional methods of K-mean and Unweighted-Pair-Group-Method-using-Arithmetic-Averages (UPGMA). The SOM method was similar to the K-means in performance, and it was the best clustering method overall because of its additional and exceptional visualization feature. Although the UPGMA method also has visualization feature and worked well with data of low complexity, its performance decreases substantially with data of medium or high complexity. |
