Description: Highlights: • Modelled fish biomass was affected by interannual variability in the plankton food. • The effects were small compared with the high variability in observations. • Fish were highly affected by changes in the larval mortality of anchovy. Abstract: The Northern Humboldt Current System is the most productive eastern boundary upwelling system, generating about 10 % of the global fish production, mainly coming from small pelagic fish. It is bottom-up and top-down affected by environmental and anthropogenic variability, such as El-Niño Southern Oscillation and fishing pressure, respectively. The high variability of small pelagic fish in this system, as well as their economic importance, call for a careful management aided by the use of end-to-end models. This type of models represent the ecosystem as a whole, from the physics, through plankton up to fish dynamics. In this study, we utilised an end-to-end model consisting of a physical–biogeochemical model (CROCO-BioEBUS) coupled one-way with an individual-based fish model (OSMOSE). We investigated how time-variability in plankton food production affects fish populations in OSMOSE and contrasted it against the sensitivity of the model to two parameters with high uncertainty: the plankton accessibility to fish and fish larval mortality. Relative interannual variability in the modelled fish is similar to plankton variability. It is, however, small compared with the high variability seen in fish observations in this productive ecosystem. In contrast, changes in larval mortality have a strong effect on anchovies. In OSMOSE, it is a common practice to scale plankton food for fish, accounting for processes that may make part of the total plankton in the water column unavailable. We suggest that this scaling should be done constant across all plankton groups when previous knowledge on the different availabilities is lacking. In addition, end-to-end modelling systems should consider environmental impacts on other biological processes such as larval mortality in order to better capture the interactions between environmental processes, plankton and fish.

Description: In geoscience and other fields, researchers use models as a simplified representation of reality. The models include processes that often rely on uncertain parameters that reduce model performance in reflecting real-world processes. The problem is commonly addressed by adapting parameter values to reach a good match between model simulations and corresponding observations. Different optimization tools have been successfully applied to address this task of model calibration. However, seeking one best value for every single model parameter might not always be optimal. For example, if model equations integrate over multiple real-world processes which cannot be fully resolved, it might be preferable to consider associated model parameters as random parameters. In this paper, a random parameter is drawn from a wide probability distribution for every singe model simulation. We developed an optimization approach that allows us to declare certain parameters random while optimizing those that are assumed to take fixed values. We designed a corresponding variant of the well known Covariance Matrix Adaption Evolution Strategy (CMA-ES). The new algorithm was applied to a global biogeochemical circulation model to quantify the impact of zooplankton mortality on the underlying biogeochemistry. Compared to the deterministic CMA-ES, our new method converges to a solution that better suits the credible range of the corresponding random parameter with less computational effort.