Extending process and understanding for the development of complex ecosystem models, with application to the Chatham Rise Atlantis model

<p>The Chatham Rise is a highly productive deep-sea ecosystem that supports numerous substantial commercial fisheries, and is therefore a likely candidate for an ecosystem based approach to fisheries management in New Zealand. This thesis describes model construction, calibration and validation, for the first end-to-end ecosystem model of the Chatham Rise, New Zealand. The work extends beyond what has previously been done for validating such models, and explores uncertainty analyses through bootstrapping the oceanographic variables, perturbing the model's initial conditions, and analysing species interaction effects, with the results further analysed with respect to known data gaps. This enables the inclusion of uncertainty in simulated scenarios using the Chatham Rise Atlantis model, thus providing an envelope of results with which to analyse and understand the likely responses of the Chatham Rise ecosystem. The model was designed with 24 dynamic polygons, 5 water column depth bins, 55 species functional groups, and used 12-hour timesteps. The transfer of energy was tracked throughout the system using nitrogen as the model's main currency. The model simulated the system from 1900–2015, preceded by a 35 year burn-in period. The model produced very similar biomass trajectories in response to historical fishing to corresponding fisheries stock assessment models for key fisheries species. Population dynamics and system interactions were considered realistic with respect to growth rates, mortality rates, diets and species group interactions. The model was found to be generally stable under perturbations to the initial conditions, with lower trophic level species groups having the most variability. The specification of the Spawning Stock Recruitment curve was explored, as it relates to the multi-species and ecosystem models within which it is now applied. Close attention needs to be given to population dynamics specific to multi-species interactions such as predation-release, in particular the Spawning Stock Recruitment curve. Potentially misleading dynamics under predation-release were identified, and the simple solution of applying a cap to recruitment when biomass exceeds virgin levels was explored. The population dynamics of myctophids under fishing induced predation release were analysed with and without limiting recruitment to virgin levels. The effects were evident in several ecosystem indicators, suggesting unintentional mis-specification could lead to erroneous model results. It raises several questions around the specification of the Spawning Stock Recruitment relationship for multispecies models, and more generally, whether the concept of ‘virgin’ (or ‘unfished’) biomass should be reconsidered to reflect dynamic natural mortality and potentially changing unfished states. The model components that had knowledge gaps and were found to most likely to influence model results were the initial conditions, oceanographic variables, and the aggregate species groups ‘seabird’ and ‘cetacean other’. It is recommended that applications of the model, such as forecasting biomasses under various fishing regimes, should include alternatives that vary these components, and present appropriate levels of uncertainty in results. Initial conditions should be perturbed, with greater variability applied to species groups modelled as biomass-pools, and age-structured species groups that have little data available from the literature.</p>