DISCUSSION 
These initial observations provide us with insight into the problems of 
running microcosms in such a way that they are analogous to some natural 
system. The natural system itself is highly variable and difficult to define, 
except within broad limits. Generally, for most of the variables measured, the 
values from the MERL microcosms fell within the ranges observed for adjacent 
Narragansett Bay. We have no evidence that the major features of 
phytoplankton and nutrient dynamics were different from Narragansett Bay. 
This lends support to the hope that the MERL microcosms will be useful 
experimental systems in which investigations will produce results transferable 
to comparable open, natural systems. 
An exception to the generalizations above was the zooplankton abundance. 
The biomass of zooplankton in the MERL microcosms (Figure 24-9) was 
somewhat less than in the bay, especially towards the end of the time period 
considered. This factor is responsible for the microcosms lying somewhat 
outside the fields for the bay data shown in Figures 24-6 and 24-8, because the 
1972-73 bay survey also returned somewhat higher zooplankton biomass 
concentrations than were found in the tanks. We believe that the tendency 
towards low zooplankton biomass was due to an artifact associated with the 
delivery of water to the tanks and that this problem will be rectified by 
subsequent changes to the plumbing. It is therefore premature now to dwell on 
the nature of the differences in zooplankton. 
The general behaviour of the nutrient and the phytoplankton data sets 
(and, to a lesser extent, that of the zooplankton) was most reassuring. No wild 
excursions occurred. The variability in the microcosms was generally similar to 
that in mid-bay stations, and the species abundances were generally similar, 
taking the data as a whole. On the other hand, the quantitative variability of 
the tanks between themselves (as may be inferred by examination of the ranges 
shown in Figure 24-3) violates our usual perception of the way in which 
experimental systems should behave. We expect them to replicate well, so that 
experiments can be performed and the results have good statistical validity. 
The difficulty with replication of nature is that nature herself is highly variable. 
Working with such systems requires that large data sets be obtained, and that 
multivariate statistical techniques be applied to reduce these correlated data 
sets to manageable formats for analysis. 
A possible way of assessing microcosm and natural system behaviour and 
developing a criterion for comparison is to calculate the generalized distance 
between data sets (Blackith and Reyment 1971). The assumptions of 
homogeneity, multivariate normality and linear correlation between variables 
must be met for such a technique to be rigorously applied but, as with all 
377 
