Allometric Trophic Network (ATN) theory of the structure and function of ecological networks is increasingly used to understand and predict complex ecological behavior including population, community and ecosystem responses to species loss, species invasion, global warming, eutrophication, fishery exploitation and evolution. Much of the success of ATN theory derives from a well-developed “metabolic theory of ecology” that easily and somewhat accurately but imprecisely estimates autotrophic growth rates, heterotrophic metabolic rates, and functional response parameters such as predators’ search and handling time of prey based on species’ body size and foraging strategy. Unfortunately, little if any work addresses how the accuracy and precision of these input parameters affect ATN model predictions. Instead, researchers typically assume more precision and accuracy of all input parameters are better which, if invalid, needlessly reduces research productivity and efficiency. We test this assumption here by systematically exploring the sensitivity of ATN models to its multiple parameters using ‘sloppy modelling’ techniques developed by systems biologists but applicable to virtually any system of differential equations. ‘Sloppy’ refers to input parameters for which large changes in their value alter model output relatively little.
Results/Conclusions
We find that ATN models express the sloppy behavior found in other models e.g., protein-protein interaction networks, in that varying most input parameters cause relatively little change in key model predictions. More specifically, we find that parameters describing most species including rare species and most species at high trophic levels cause little change in the overall biomass of the whole ecosystem. Instead, abundant plants’ carrying capacity and abundant animals’ (especially abundant animals with few consumers) metabolic rates, maximum consumption rates, and Hill coefficients of their functional response appear to be the critically important “stiff” parameters for which small variations cause relatively large changes in model output. When evenly weighing variation of all species’ abundances, Hill coefficients for rare and the most abundant (but not intermediately abundant) animals are relatively stiff as are the maximum consumption rates of abundant animals while model variation is remarkably sloppy with respect to plant growth rates and carrying capacities. Half saturation densities and especially predatory interference/prey defense parameters of the functional response are surprisingly sloppy for both overall ecosystem and all species’ responses. These findings suggest that focusing estimation effort toward the few stiff parameters and away from the many sloppy parameters within the context of specific research objectives can greatly increase the productivity and efficiency of predicting ecological systems.