Date of Award
Open Access Dissertation
John A Kupfer
Ecosystem structure and function are the product of biological and ecological elements and their connections and interactions. Understanding structure and process in ecosystems is critical to ecological studies. Ecological networks, based on simple concepts in which biological and ecological elements are depicted as nodes with relationships between them described as links, have been recognized as a valuable means of clarifying the relationship between structures and process in ecosystems. Ecological network analysis has benefited from the advancement of techniques in social science, computer science, and mathematics, but attention must be paid to whether the designs of these techniques follow ecological principles and produce results that are ecologically meaningful and interpretable. The objective of this dissertation is to examine the suitability of these methods for various applications addressing different ecological concerns. Specifically, the studies that comprise this dissertation test methods that reveal the structure of various ecological networks by decomposing networks of interest into groups of nodes or aggregating nodes into groups. The key findings in each specific application are summarized below.
In the first paper, REgionalization with Clustering And Partitioning (GraphRECAP) (Guo 2009) and Girvan and Newman's method (Girvan and Newman 2002) were compared in the study of finding compartments in the habitat network of ring-tailed lemurs (Lemur catta). The compartments are groups of nodes in which lemur movements are more prevalent among the groups than across the groups. GraphRECAP found compartments with a larger minimum number of habitat patches in compartments. These compartments are considered to be more robust to local extinctions because they had stronger within-compartment dispersal, greater traversability, and more alternative routes for escape from disturbance. The potential defect of the Girvan and Newman's method, an unbalanced partitioning of graphs under certain circumstances, was believed to account for its lower performance. In the second study, Modularity based Hierarchical Region Discovery (MHRD) and Edge ratio-based Hierarchical Region Discovery (EHRD) were used to detect movement patterns in trajectories of 34 cattle (Bos taurus), 30 mule deer (Odocoileus hemionus), and 38 elk (Cervus elaphus) tracked by an Automated Telemetry at Starkey National Forest, in northeastern Oregon, USA. Both methods treated animal trajectories as a spatial and ecological graph, regionalized the graph such that animals have more movement within the regions than across the regions, and then investigated the movement patterns on the basis of regions. EHRD identified regions that more effectively captured the characteristics of different species movement than MHRD. Clusters of trajectories identified by EHRD had higher cohesion within clusters and better separation between clusters on the basis of attributes of trajectories extracted from the regions. The regions detected by EHRD also served as more effective predictors for classifying trajectories of different species, achieving a higher classification accuracy with more simplicity. EHRD had better performance, because it did not rely on the null model that MHRD compared to, but invalid in this application.
In the third study, a proposed Extended Additive Jaccard Similarity index (EAJS) overcame the weakness of the Additive Jaccard Similarity index (AJS) (Yodzis and Winemiller 1999) in the aggregation of species for the mammalian food web in the Serengeti ecosystem. As compared to AJS, the use of the EAJS captured the similarity between species that have equivalent trophic roles. Clusters grouped using EAJS showed higher trophic similarities between species within clusters and stronger separation between species across clusters as compared to AJS. The EAJS clusters also exhibited patterns related to habitat structure of plants and network topology associated with animal weights. The consideration of species feeding relations at a broader scale (i.e., not limited in adjacent trophic levels) accounted for the advantages of EAJS over AJS. The concluding chapter summarizes how the methods examined in the previous chapters perform in different ecological applications and examines the designs of these algorithms and whether the designs make ecological sense. It then provides valuable suggestions on the selections of methods to answer different ecological questions in practice and on the development and improvement of more ecological-oriented techniques.
Gao, P.(2013). Uncovering the Structures In Ecological Networks: Multiple Techniques For Multiple Purposes. (Doctoral dissertation). Retrieved from https://scholarcommons.sc.edu/etd/2567