EECS Publication
Combinatorial Algorithms and High Performance Implementations for Elucidating Complex Ecosystem Relationships from North Sea Historical Data
M. A. Langston, A. D. Perkins, D. J. Beare, R. W. Gauldie, P. J. Kershaw, J. B. Reid, K. Winpenny and A. J. Kenny
This investigation centers on elucidating complex relationships among quantifiable variables of significance to the North Sea ecosystem. These variables encompass a huge variety of biotic and abiotic factors, and tend to possess divergent periodicities and other diverse properties. Novel mathematical tools and powerful graph algorithms are described that can be harnessed to uncover temporal, spatial and other meaningful relationships on an immense scale. High performance parallel implementations can be synthesized to extract and highlight variable sets common to multiple relationships (cliques), and to determine inflection points, putative regime changes and other patterns of possible interest. These approaches are discussed in the context of more traditional clustering methods. Data quality and the significance of missing or corrupted values are also addressed, as is the importance of mining data at multiple levels of granularity. A long-term goal is to establish data dependencies upon which we can draw conclusions about the impact of man and other agents upon the Sea.
Published 2007-02-01 05:00:00 as ut-cs-07-590 (ID:112)