Towards a strategic framework for assessing uncertainty in ecosystem dynamics models: objectives are sensitive too

CCAMLR’s working group on Ecosystem Monitoring and Management intends to evaluate candidate feedback management approaches for the Antarctic krill fishery in the southern Drake Passage and Scotia Sea. The Foosa ecosystem dynamics model was developed to perform spatially resolved, stochastic simulations of krill, its predators and fishery. It has been used to provide advice on the spatial allocation of the precautionary catch limit for krill. Foosa might therefore be a suitable simulation platform for evaluating feedback management approaches. Although Foosa resembles a minimum realistic model, it has 50 categories of input parameter and the four parameterisations used to provide spatial allocation advice each required values for 2,311 distinct inputs. In the absence of guidance about management reference points, the modellers provided advice in terms of 41 illustrative reference points intended to represent the key objectives of ecosystem based management: ecosystem productivity, health, resilience and services. Foosa was developed and evaluated through iterative interaction with scientific working groups. There is a global need to develop strategic frameworks for assessing uncertainty in ecosystem dynamics models. Such a framework could incorporate elements of the evaluation of Foosa, but it should also include analysis of the sensitivity of model outputs to model inputs. Here we provide a local sensitivity analysis for Foosa. Some of the outputs used to provide advice on the spatial allocation of krill catch were insensitive to all perturbations, whereas the response of others was upto 8 times the perturbation. They were most sensitive to a parameters controlling predator recruitment through stock recruit relationships and pre-recruit mortality.  A parameter mediating the effect of a forcing function on krill recruitment, which was used to condition the model on historical dynamics, was also important. It is apparent from our analysis that sensitivity is as much a function of model outputs as inputs, which suggests that indentifying quantitative reference points for ecosystem based management will be an essential part of any strategic framework for assessing uncertainty.