Tosca Ballerini: Towards good practices in end-to-end modeling

The development of ecosystem models is often accompanied by the aspiration to use them for practical purposes such as evaluating the risk associated with alternative fishery management options. To move toward this goal, ecosystem models need to take into account the uncertainty associated with the data, the model structure, and the modeling assumptions. We show how these three factors affect our ability to identify differences in the behavior of similarly structured end-to-end models for two Antarctic marine ecosystems built using the ECOTRAN routine. For each region, systematic sensitivity analyses in every trophic linkage were performed and ecosystem metrics were calculated from 1000 random thermodynamically balanced ecosystems. These metrics included the efficiency of energy transfer from primary producers to top predators and a measure of each group’s importance as an energy transfer node. We found that in both systems a change in primary production leads to a decline in upper trophic level biomass, since the upper trophic level groups are sensitive to the amount of biomass channeled through the fish groups. The two original models, however, made different assumptions about the fish groups, and while both sets of assumptions are plausible, any practical application of the predictions requires an awareness of the implications of the original assumptions. Our modeling exercise shows that predictions from end-to-end models are highly affected by multiple sources of uncertainty. There is a need to improve the quality of the data and process assumptions used to implement the models, and there is also a need to identify management strategies that are robust to uncertainty, i.e. we need to make sure that we are not using management approaches that require more accuracy than models can provide. The standards developed for general circulation models, which are used for weather forecasting and climate projection, provide a reference point for developing standards and good practice in ecosystem models. Our approach implements some of this good practice. In order to be useful for practical applications and provide robust information to management, we recommend that end-to-end models include explicit evaluation of uncertainty; that the modeling assumptions are clearly stated and their implications are discussed; that the data and the code used to implement the models are open so that verification and improvements can be made by the larger end-to-end modeling community.