##
COS 75-9
Statistical analysis of sixty system-wide ecological network measures

**Background/Question/Methods**

** **Ecosystems are often described as flows among compartments, which form ecological networks. Over sixty model based ecosystem measures (or indicators, properties, indices) have been proposed, claiming to quantify various aspects of ecosystem state, structure, or function. Some of these measures are purely structure based (e.g. link density), some utilize flow rates among compartments (e.g. cycling index), and some utilize thermodynamic principles (e.g. ascendancy). There exist multiple software (e.g. Ecopath, EcoNet) and packages (enaR) that compute these measures for a given appropriate ecosystem model.

Each of these over sixty systemwide measures provide information about a different aspect of the ecosystem model. However, within the set of feasible ecosystem models, how different or independent is the information provided by a given measure, compared to the remaining sixty or so measures? In other words, does a specific measure really provide any new information, or can it be expressed as a function of the remaining measures? What is the dimension of the actual information space provided by all these measures?

We looked for answers to these questions by computing all measures for 52 real life ecosystem models collected from published literature, and analyzing the data using various statistical techniques.

**Results/Conclusions**

** **While the chosen statistical method, and the parameters within a given method changes the results slightly, we observed that the actual dimension of the sixty ecosystem measures range between 9 and 15. In other words, significantly less measures provide equivalent information about a given ecosystem model. Some measures were over 95% correlated. Cluster analysis provided groups of closely correlated measures.

We observed that some measures have stronger correlation than expected, such as (e.g. ascendency, total system throughput, average path length) and some measures may not have strong correlations as previously reported, such as the relation between indirect effects with network size, or connectance and total system throughflow.

This comprehensive analysis of measures helps us assess the actual new information provided by each additional measure, within the space of actual real life ecosystem models, as opposed to a specific class of models, or hypothetical systems.