Structuring of ecological data sets by methods of correlation and cluster analysis

Abstract

The analysis of ecological field data of extensive studies in marine environments needs special adapted strategies for their analyses. Those data sets can be classified into two major categories: (a) data from stations, i.e. quasi-synoptic approaches for the description of different biotopes; and (b) data from seasonal investigations, i.e. surveys of time dependent changes in an environment. The unbiased search for the various interrelationships between parameters becomes a useful tool in describing ecological patterns of complex systems. The evaluation of main items by cluster analysis and other ordering procedures defines structural and functional units or sets local and/or seasonal boundaries. It is proposed to use non-parametric statistics to overcome the inhomogeneity of the parameters and their interactions which may be assumed to be nonlinear or governed by many feed-back mechanisms. Subsequent analysis of those correlation matrices can be conducted by different cluster algorithms and shows in a holistic view on the original data set significant patterns of the structure of the system and leads to the description of interacting subsystems at different levels. Such unbiased data analysis has been adopted to various ecological data sets comprising physical, chemical and biological parameters. The analysis of data from seasonal and synoptic investigations shows the suitability of the strategy proposed: (1) An annual cycle of data concerning the microbial environment in the Kiel Fjord (Baltic Sea) revealed distinct seasonal patterns which can be attributed to the general planktonic environment. It is possible to define subunits of the total annual cycle, which may be regarded as ‘ecological situations’, their successions and boundaries based on thresholds of environmental parameters. These thresholds are regarded as ‘switches’ of the ecosystem's structure or as marginal conditions of the acting subsystem. (2) The analysis of data sets from different stations showed the separation of a ‘unique’ environment into units of different structures, caused by influences of the current system or other dominating environmental conditions. These results point to a strong spatial heterogeneity of marine systems, indicate their response on different nutritional sources and reflect the change in the interrelationships. The analysis of the interacting or residual (unclustered) parameters of the subunits either from seasonal of spatial data sets shows the different structures of these systems as for instance the changing ratios of nutrients, populations and environmental characters. The lack of causal relationships between individual parameters of complex systems and the stochastic behaviour of their course leads to the assumption that the annual cycle of the planktonic environment is more governed by catastrophes, i.e. sudden and unpredictable events, than by continuous processes. The different states of the system as established by this method show the fluctuation of parameters according to their marginal conditions and their ranges of variability within the given situation. The effects of sudden changes in environmental conditions were tested in bioassays using containers of approximately 30 1 of waters. The results reflect the response of the microbial population and their activity on various impacts. Thus it becomes more evident to focus on features describing more overall patterns of an ecosystem in terms of stability or instability and boundary conditions than on direct interactions between individual parameters based on the concept of cause and effect.