Changing climates: Unlocking the past to improve predictions of the future
Background
Equilibrium climate sensitivity (ECS) is the long-term global mean temperature increase expected from a doubling of atmospheric CO₂ concentration compared to pre-industrial levels. It is a valuable metric for understanding future climate change and associated risks. But, ECS remains highly uncertain; the World Climate Research Programme estimates it to be somewhere between 2.3 and 4.5°C.
The only observational records of what climate was like under vastly different levels of atmospheric CO₂ come from Earth’s past, such as the cold Last Glacial Maximum or the warm Pliocene. Information about those climates can be combined with the complex physics represented in models using an approach that accounts for the full uncertainty in climate simulations to understand & robustly quantify ECS, in particular constraining the ECS upper limit. This helps inform policy makers on the risk and likelihood of 21st century climate change being at the high end of ECS and climate projection uncertainty.
PhD Opportunity
Using advanced uncertainty quantification, this project will use machine learning to perform ensembles of simulations (Hourdin et al., 2023) from state-of-the-art climate models and combine these with observational records from Earth’s past, providing ‘out of sample’ constraints on the climate response to changing levels of atmospheric CO2.
A coexchangeable process model (Astfalck et al., 2024) will be used to capture multiple correlated spatio-temporal fields of climate variables, merging many simulations of the past with corresponding records of sea surface conditions, to produce an ensemble of physically-consistent, observation-constrained sea surface temperature and sea-ice maps. These surface conditions will be used to drive climate simulations performed with the UK flagship atmosphere General Circulation Model, HadGEM3-GC5c, in collaboration with the UK Met Office (CASE partnership). From the new simulations, state-of-the-art estimates of ECS will be derived, expanding the methods of Cooper et al. (2024) to include robust quantification of uncertainties from sea surface patterns, atmospheric parameters and ice sheet reconstructions. The work will be able to answer exciting and important research questions, such as:
What plausible patterns of surface climate can be produced for times in the past when atmospheric CO2 was different to today?
What are the levers of uncertainty (in models and observations) that affect those patterns, and how big are the effects?
What do the results tell us about climate feedbacks and ECS, and thus future climate change? Does such a robust consideration of the various sources of uncertainty reduce the current range in ECS and 21st century warming?
Applicant Profile
The project would suit students from Physics, Mathematics, Oceanography, Meteorology, Climate Sciences, Natural Sciences, Earth/Environmental/Geographical Sciences. Experience in computer programming (e.g., Python, Fortran, C++, MATLAB, R…) or numerical modelling is highly desirable.
Other information
Hourdin, F., et al. Toward machine-assisted tuning avoiding the underestimation of uncertainty in climate change projections. Science Advances 9, eadf2758 (2023).
Cooper, V. T. et al. Last Glacial Maximum pattern effects reduce climate sensitivity estimates. Science Advances 10, eadk9461 (2024).
Astfalck, L., et al. Coexchangeable Process Modeling for Uncertainty Quantification in Joint Climate Reconstruction. Journal of the American Statistical Association 119, 1751–1764 (2024).
