Ocean Biogeochemical Optimisation in Earth System Models

Lead Research Organisation: University of Oxford
Department Name: Environmental Research DTP

Abstract

The key aim of my research project is to reduce the uncertainty in projections of future climate change. I will be improving the next generation Met Office/NERC Earth System model (UKESM1) which is currently under development. This project is in collaboration with the Earth Sciences and Mathematics departments of Oxford University, ocean biogeochemists at the National Oceanography Centre in Southampton, and an Earth system modeller at the Met Office.
Global climate change predictions from the numerous computational models show a spread in the predicted warming associated with the rising atmospheric carbon dioxide concentrations. This large uncertainty makes it difficult to a) convince, and b) advise policy makers of global warming and how best to tackle it. We must reduce the uncertainty of these earth system models by better understanding the various processes which occur and better parameterising them within the models. This will act to "hone in" on the actual degree of global warming we will experience in the future.
The ocean is a great sink of carbon dioxide and has absorbed much of the carbon dioxide emitted by humans since the industrial revolution. The ocean does this primarily due to the high solubility of CO2 in seawater, as well as complex interactions between nutrients, plankton and carbon. All of these interactions are simulated by a model called MEDUSA (Model of Ecosystem Dynamics, nutrient Utilisation, Sequestration and Acidification) existing within UKESM1. It does this through a range of parameterisations that include a number of key uncertain parameters, which I will work to improve. This will be particularly challenging due to the complex interaction between biogeochemistry and ocean circulation, which cause the model sensitivities to vary through both space and time. Also, to be able to systematically optimise the performance of a model as computationally expensive as MEDUSA, I will have to exploit recently-developed mathematical optimisation techniques.
I aim to accomplish 3 main goals:
1. Estimate MEDUSA's sensitivity to various parameters and thus the relative importance of key ocean biogeochemical processes that affect the sequestration of carbon in the ocean. To do this I'll systematically vary the individual parameters and compare the relative changes in CO2 drawdown by the oceans.
2. Attain an optimal set of MEDUSA's parameters that minimizes the misfit between observations and the model.
3. Carry out a quantitative assessment of the impact of parameter optimisation on key aspects of UKESM1-projections, such as global climate sensitivity, marine carbon uptake and the resulting biogeochemical state of the ocean. To attain this, I'll compare the UKESM1 results between the pre-optimised MEDUSA and post-optimised MEDUSA.

Publications

10 25 50
 
Description Two billion times acceleration of scientific simulations with deep neural architecture search 
Organisation University of Oxford
Department Department of Physics
Country United Kingdom 
Sector Academic/University 
PI Contribution I provided training data for a research to use as a case study for training a neural network, and I wrote 2 paragraphs to go in their scientific paper (currently in review), on which I and my primary supervisor are named as co-authors.
Collaborator Contribution N/A
Impact The outcome is a multi-disciplinary scientific paper (in review), involving the disciplines physics, earth sciences and geology.
Start Year 2019
 
Title OptClimSO 
Description This code package allows any parameterised computational model to be optimised by any numerical optimisation algorithm, originally written by Simon Test (University of Edinburgh), which I have adapted and updated to allow for optimisation of a specific ocean biogeochemical model and several optimisation algorithms (see https://github.com/SophyOliver/OptClim/blob/master/OptCLim_UserGuide.pdf). 
Type Of Technology Software 
Year Produced 2019 
Impact Used by me in my research, and by ETH Zurich. 
URL https://github.com/SophyOliver/OptClim