This week’s insightful analysis is for Naeem et al’s sweet 1995 paper on their ecosystems-in-boxes experiment at the ecotron, in which they manipulated species diversity in producer-consumer-predator systems and measured ecosystem functions for 200 days.

The original paper: Empirical Evidence that Declining Species Diversity May Alter the Performance of Terrestrial Ecosystems, Philosophical Transactions of the Royal Society: Biology, 347(1321) 1995.

In general, I found this experiment compelling and well-formulated. Although the implications of the results are necessarily limited by the scope and structure of the experiment, it represents new information and is suggestive of further experiments to come. The complaints below are just complaints.

First, an editing concern: I found it difficult to figure just how many replicates of each treatment were conducted. This is important information that could have been highlighted. If I read this wrong (14 replicates of 4 treatments all over one ~200 day run?) then several of my further concerns could be baseless.

The size of the experiments (2m x 2m?) may have been too small to allow for spatial heterogeneity and patchiness, which may be an important factor in real ecosystems. Small and well-mixed box and jar experiments seem more likely to behave according to simpler classical equation-based ecology. Larger experiments which allow for predator-prey disruption and locality of form and function can disrupt those simple relationships and introduce non-linearity and randomness, which may need to be represented for an experiment to analogous to actual ecosystems. In this case the species were small (molluscs and such) so perhaps 4 square meters was sufficient for some spatial effects, but it isn’t addressed in the paper.

The number of experiments (if I read it right) was awfully limited. If there were any non-linearities or sensitivity to initial condition emerging from the aggregation of multiple similar species, then different runs could potentially yield very different results. Controlling for that would require many more replicates. Just finding out if it needed to be controlled for would require several more replicates.

There were several assertions of expertise on the experimental design: for instance, the choice of species (based on a “ruderal British” system) wasn’t justified. The species chosen were “ecologically similar in their biology and capable of growing in the ecotron”. That a priori assertion of ecological similarity may represent either a tautology in the experiment if it’s right, or a weakness in experimental design if it’s wrong, depending how you look at it.

It’s not entirely clear if equilibrium was reached during the 200 odd days of the experiment, or what the form of the equilibrium would have been were it reached (flat-line, switching off of dominants, something more complicated). Would the results of the experiment at 200 days have seemed as compelling if, after another 100 days, several of the chambers experienced ecosystem failure? If equilibrium is not a common feature of real ecosystems, perhaps they would. A longer time sequence to explore if the trends in the ecosystems were cyclical, linear or random-ish would have helped frame the relevancy of the results at the 200th day. The authors assert that they “believe that longer running times would only have allowed the different treatments to diverge further in their different levels of expression of ecosystem processes”. That is a strong claim. And any purely linear character of an ecosystem, run out indefinitely seems likely to end in a crash.