An Exploratory Analysis of Microcosm and Ecosystem Behavior Using Multivariate Techniques

Abstract

A conservative data set was chosen to describe ecosystem behavior: total zooplankton (mg dry wt m-3), chlorophyll a (W I-'), ammonia, nitrite, plus nitrate, phosphate and silicate (pg-at I-'). These variables were measured weekly in the microcosms, as well as in a field survey of Narragansett Bay in 1972-73; thus a comparison data set was available. Two types of statistical techniques were used to explore the question of divergence among replicate microcosms and treatment groups (including the bay stations as treatment groups): Multiple Discriminant Analysis and Distance Statistics. The first was used to explore the major axes of variation between replicate tanks and bay stations. Distance statistics were used to describe the divergence of replicate tanks and treatment groups as a function of time. The first distance analysis was based on discriminant space using the pooled covariance matrix and the distances were weighed by the percent variation explained by the axes. The second distance analysis did not rely on discriminant space and the distances were based on estimations of the individual covariance matrices rather than the pooled covariance matrix. Discriminant analyses were en~ployed to explore the major axes of variations between microcosm treatments and the natural system. Generally only the first axis was interpretable In the microcosm perturbation experiment, the first axis was interpreted to be indicative of bloom conditions. When the microcosms and the bay were analysed, the first axis was interpreted as an eutrophication gradient. In addition, time varying plots were made to indicate differences due to treatment and season. The two distances indicated obvious features in the divergence among replicate tanks and within bay areas and between treatment groups and bay areas. With both distance statistics, the divergence among replicate tanks within a treatment could be related to lag effects in the dynamics of plankton blooms. Divergence among treatment groups could b e related to substantial changes in the functional behavior of the microcosms. The natural logarithm of the determinant of individual covariance provided evidence that the bay stations were more variable than the microcosms. The question remains as to whether the techniques we have used are more useful than visually interpreting plots of the response of the individual variables over time. The attempts to reduce the dimensionality of the representation space did provide a summarization of the data and resulted in interpretations in agreement with raw data time series analyses The distance metrics have roughly quantified the magnitude of divergence in microcosm behavior and compared the behavior with the natural system.