Ocean biogeochemical optimisation in Earth System Models

Thesis

Global ocean biogeochemical models are important tools to understand the cycling of nutrients and carbon in the global ocean, and when embedded within Earth System Models allow us to investigate how the oceans interact in the past, present and future with the entire earth system. Biogeochemical parameters within such models can be systematically calibrated by minimising the misfit between modelled tracers and observations, yet due to their large computational expense this is rarely carried out. In this thesis, the aim is to first investigate efficient ways to tune global ocean biogeochemical parameters using optimisation algorithms, so to encourage more frequent objective model improvement. We then apply them to the global ocean biogeochemical model MEDUSA (Model of Ecosystem Dynamics, nutrient Utilisation, Sequestration and Acidification), the marine biogeochemical component of the U.K. Earth System Model (UKESM1) .

First, we carry out a parameter optimisation and sensitivity study on the global ocean biogeochemical model MOPS (Model of Oceanic Pelagic Stoichiometry). We investigate the performance of DFO-LS (Derivative-Free Optimisation using Least Squares), an optimisation algorithm which has been designed for computationally expensive models, and compare it to CMA-ES (Covariance Matrix Adaptation Evolution Strategy), another algorithm that was applied in a previous study in one of the first successful attempts at calibrating a global model. DFO-LS successfully calibrated MOPS in less than 1/30th of the computational expense required by CMA- ES, making it feasible to move on to calibrating more computationally expensive models.

We then investigate the influence on parameter optimisation of shortening the length of the ocean biogeochemical spin up, which is required for the model to reach an equilibrated state. If this can be reduced from over 3000 years during optimisation, then the whole process will become less computationally expensive. Unfortunately we show for MOPS the spin up could only be reduced by one third, to a minimum length of 2000 years, due to the influence of certain biogeochemical parameters on the deeper ocean, where timescales are longer.

Lastly, parameter sensitivity and optimisation studies are carried out on the more complex model MEDUSA. We provide an assessment of how different biogeochemical parameters influence global distributions of modelled oceanic variables, and find the most influential parameters to be those which control diatom and meso-zooplankton abundance, as these heavily influence the transport of organic matter to the deeper ocean. We also provide several optimised configurations of the model with a closer fit to observed nutrients and oxygen than the current default configuration, however with some unrealistic modelled primary production and plankton abundance.

Authors: Oliver, S

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: Parameter optimisation; Ocean biogeochemical modelling
• Subjects: Ocean biogeochemistry; Marine sciences; Numerical optimisation