About the Project
Background
Climate models, despite ongoing improvements, contain uncertainties and biases, whether fully process-based, AI-driven, or a hybrid of the two. For example, one persistent source of uncertainty in the climate system is the land surface component, with a wide range of possible futures across climate models as to how much CO2 will be taken up by the land under climate change, and whether this will be a carbon source or sink. For models to be useful for predicting future climate, these types of…
uncertainties need to be understood and constrained or quantified.
There are many drivers of model uncertainty. Simulations can be computationally expensive, and statistical-machine learning 'emulators' are commonly used as fast approximations, allowing more rigorous exploration of the high-dimensional input and output spaces, and enabling calibration of inputs to be more feasible. This project will build on past work calibrating environmental models to help reduce uncertainty in future climate.
PhD Opportunity
The overarching aims of the project are to develop statistical, machine learning or AI uncertainty quantification (UQ) methodologies to enhance and enable efficient climate model calibration, advancing the standard approaches of Gaussian Processes and history matching. The project will combine the core UQ tasks of model emulation and calibration, and within each there are many avenues that can be developed according to a student's interests:
- Emulation: across large numbers of unknown parameters, across boundary conditions, across future emissions scenarios, or all these unknowns at once. Training emulators in large input spaces and across multiple related outputs is non-trivial and may require novel methodology. Training emulators to coupled models is challenging due to computational expense but may be achievable via multi-level emulation techniques.
- Calibration: history matching is a powerful technique to identify 'good' versions of model inputs, but which outputs we emulate and calibrate to, and how structural error (differences between model-world and reality that can't be removed by tuning) is defined affect the results. This project could explore and develop techniques for handling structural uncertainty more rigorously in the calibration process.
Regardless of the route taken, this project will develop methods that will have the potential to enhance the understanding of uncertainty across a range of climate applications, for example feeding into better understanding and constraint of uncertainties and sensitivities relating to future projections of the carbon cycle.
The project will be led by Dr James Salter (Exeter) with co-supervision from Dr Victoria Volodina, both of whom have extensive experience developing and applying UQ techniques in environmental modelling. Through the project there will be opportunities for collaborations with climate modelling experts and to work on cutting-edge problems at the intersection of climate modelling and UQ.
Applicant Profile
Students with a strong background in mathematics, statistics, computer science or another quantitative subject, with an interest in climate or environmental problems, are encouraged to apply. Experience working with climate model data is not necessary but is beneficial.
Funding Notes
This project is part of the UNRISK CDT, which offers 15-18 fully-funded NERC studentships, covering full university tuition fees; a personal stipend at standard UKRI rates; £6000 individual research and training costs; £5000 (per student) of cohort-level training; and a 'Flexible Fund' for special projects.
International applicants will need to cover costs related to applying for a student visa and the international health surcharge (IHS)
Applications are open to UK and international applicants. The number of awards for international applicants is limited by UKRI rules.
More information is available on the UNRISK website: View Website
References
Emulation/calibration: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2020MS002217
Perturbed parameter ensembles: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4341/