Increasing the reliability of the Bay of Biscay Atlantis model: A sensitivity analysis to parameters perturbations using a Morris screening approach

Abstract

The Bay of Biscay Atlantis model is expected to become a significant tool for exploring how anthropogenic impacts and environmental variability, in particular fisheries and climate change, could affect the Bay of Biscay ecosystem. Before being used for this purpose, we wanted to identify and evaluate sensitive parameters. Here we present the results of a sensitivity analysis of growth, recruitment, predator-prey availability and mortality parameters for fourteen key or sensitive species/functional groups that make up the food-web of the entire ecosystem, while keeping environmental parameters constant. Specifically, a Morris screening approach has been used to manage the number of parameters and species/functional groups we want to analyse given the computational cost of the simulations. With a small number of simulations, this method can efficiently provide information on main parameter effects and interactions, including identifying non-linear effects. The sensitivity analysis showed that maximum growth rate of large phytoplankton and macrozooplankton, as well as predator-prey availability were the most sensitive parameters. The effects were found to be monotonic or almost-monotonic except predator-prey availability of mesozooplankton, which was non-linear and/or with interactions with other inputs parameters. The responses to the perturbations varied by trophic level. Overall, top predators and fish species/functional groups responded weakly to parameter perturbations, except pelagic shark, baleen whales, toothed cetaceans, horse mackerel, anglerfish, blue whiting, hake, and megrim. Pelagic and benthic invertebrate species/functional groups and primary producers were the most vulnerable to perturbations. The information gained from this sensitivity analysis provided a better understanding of the model structure and highlighted the importance of bottom-up effects.