Evidence for resource homogenization in 50 trophic ecosystem networks

Connectivity patterns of ecological elements are often the core concern of ecologists working at multiple levels of organization (e.g., populations, communities, ecosystems, and landscapes) because these patterns often reflect the forces shaping the system’s development as well as constraining their operation. One reason these patterns of direct connections are critical is that they establish the pathways through which elements influence each other indirectly. Here, we tested a hypothesized consequence of connectivity in ecosystems: the homogenization of resource distributions in flow networks. Specifically, we tested the generality of the systems ecology hypothesis of resource homogenization in 50 empirically derived trophic ecosystem models representing 35 distinct ecosystems. We applied Network Environ Analysis (NEA) to calculate resource homogenization for these models, where homogenization is defined as the ratio of the coefficient of variation of the direct flow intensity matrix (CV(G)) to the covariance of the integral flow intensity matrix (CV(N)). A ratio greater than unity indicates the presence of homogenization. We also tested the hypotheses that homogenization increases with system size, connectance, and cycling. We further evaluated the robustness of our results in two ways. First, we verified the close correspondence between the input- and output-oriented homogenization values to ensure that our results were not biased by our decision to focus on the output orientation. Second, we conducted a Monte Carlo based uncertainty analysis to determine the robustness of our results to ±5%±5% error introduced into the original flow matrices for each model. Our results show that resource homogenization occurs universally in the 50 ecosystem models tested, with values ranging from 1.04 to 1.97 and a median of 1.61. However, our results do not support the hypothesized relationship between network homogenization and system size and connectance, as the results of the linear regressions are insignificant. Further, there is only weak support for the positive relationship between homogenization and cycling. We confirm that our results are not biased by using the output-oriented homogenization values instead of the input-oriented values because there is a significant linear regression between the two types of homogenization (r2=0.38r2=0.38, p<0.001p<0.001) and the values are well correlated (S=8,054S=8,054, ρ=0.61ρ=0.61, p<0.001p<0.001). Finally, we found that our results are robust to ±5%±5% error in the flow matrices. The error in the homogenization values was less than the error introduced into the models and ranged from a minimum of 0.24% to a maximum of 1.5% with a median value of 0.58%. The error did not change the qualitative interpretation of the homogenization values. In conclusion, we found strong support for the resource homogenization hypothesis in 50 empirically derived ecosystem models.