Changes in the global climate caused by human activity are already affecting many weather and climate extremes in every region worldwide. Under current policies, it is estimated that we can expect to see 2.7°C of global warming by 2100. With a warming climate, we can also expect to see an increase more extreme, or black swan events as heatwaves, flooding, droughts, crop failures, wildfires and tropical cyclones. A recent study has shown that 3-16% of people born between 2003 and 2020 will experience unprecedented lifetime exposure to such events that would not occur if global warming were kept to 1.5 °C. Although the scientific consensus is that the climate will continue to change and have worsening effects on human life, the details of predictions, such as the rate of change, can be more uncertain due to the complex nature of the Earth Climate System. Therefore, the challenge lies in mapping model forecasts of possible future climates in robust ways that can capture the likelihood of extreme events, thereby informing mitigation and adaptation planning.
Currently, the method used to achieve this involves downscaling coarse-resolution climate models with Numerical Weather Prediction (NWP) models for regional climate estimates. However, one disadvantage of this approach is that it adds additional high overheads to the computational costs of traditional climate modelling. Recent advances in NWP emulators (e.g., AIFS from ECMWF, GraphCast from Google) combine Artificial Intelligence with other data science developments as an alternative to this approach. Once trained, these new technologies have the potential to reduce the carbon footprints of these downscaling activities. However, early testing of these AI weather forecast models has revealed that although they perform better in terms of standard scoring, they produce less detailed physical output. For example, common metrics obscure poor performance around phenomena such as category five hurricanes, where these models lack out-of-distribution generalisation (extrapolation) for extreme weather events (e.g. Figure 1). Therefore, to realise the full potential of these new AI tools for climate adaptation and mitigation studies we need to improve their performance such that they can predict black swan events.
Figure 1: Satellite view of Category 5 Hurricane Erin, the first hurricane of the 2025 as it moves across the Atlantic. Since August 7th NOAA, the organisation in America responsible for tracking weather from space, has updated its prediction that 2025 will be an above normal seasonal outlook. Therefore, we are set to see more storm with the strength of Erin in the months ahead.
This project is a CENTA Flagship Project.
This project is suitable for CASE funding
Each host has a slightly different application process.
Find out how to apply for this studentship.
All applications must include the CENTA application form.
Choose your application route
In this project the student will learn to work with the new generation of NWP emulators (NWPE) and initially investigate:
From here the student would then focus on developing new data driven approaches from predicting extremes, starting off with simpler concept models and building them into robust frameworks. A component of this work will involve exploiting novel observational datasets that have not yet been incorporated into these types of study. Finally, this project will also track advances the current (7th) Coupled Model Intercomparison Project (CMIP7) and seek to compliment or find new collaborative partnerships to advance the study.
DRs will be awarded CENTA Training Credits (CTCs) for participation in CENTA-provided and ‘free choice’ external training. One CTC can be earned per 3 hours training, and DRs must accrue 100 CTCs across the three and a half years of their PhD.
NCEO will provide access to its Researcher Forum, staff conferences/workshops and national-level training. This has included courses machine learning and data assimilation.
The UK Met Office will provide training and the opportunity to work alongside scientists working on climate extremes. This includes spending dedicated time in Exter with the HadEX team.
University of Leicester will provide training on using the energy balance model and data processing on the ALICE (Leicester) and JASMIN (NERC) HPC facilities. The student will also have the opportunity to take suitable modules from the UG and MSc courses in climate physics, satellite data analysis and Python.
Established in 1990, the Met Office Hadley Centre is one of the UK’s foremost climate change research centres. Dr Robert Dunn leads and been involved in the development of several underpinning in situ datasets for detecting climate change and climate extremes. In addition to hosting student placements, MOHC will also provide access to and support with flagship datasets. Through, this project the student will also be introduced to a broader network through the MOHC Climate Monitoring group and AI4Climate.
Throughout the entirety of this project the student will spend some time at MOHC each project year, the duration of each visit to be discussed and decided upon once the project is underway to ensure appropriate support.
Year 1: Review of existing literature, refinement of research plan & objectives, and training. Initial training will focus on computing skills needed for the project, including (but not restricted to) coding, HPC usage, executing model runs, and attendance at 1-2 suitable meetings/workshops. Analysis of NWPE output from freely existing models with climate quality in situ and satellite products will allow the student to refine the research plan and objectives. The final objective for the year would be for the student to feedback findings to UKMO through AI4Climate and present their work at an appropriate national/international conference.
Year 2: Design, run and analyse current ML experiment(s) for predicting extreme cases outside traditional training datasets with the objective to publish the results in a peer-reviewed journal. The student would also look to present at one international meeting/conference and one national conference during this year.
Year 3: Consolidation of work from the first two years to run experiments focusing on future projections from CMIP7. The exact nature of this work would be reviewed before starting the third year of the project. The student would also look to present at one international meeting/conference and one national conference during this year and publish the study in a peer reviewed journal.
Plésiat, É., Dunn, R.J., Donat, M. and Kadow, C., 2025. Reconstructing Historical Climate Data using Deep Learning (No. EGU25-18816). Copernicus Meetings.
Sun, Y.Q., Hassanzadeh, P., Zand, M., Chattopadhyay, A., Weare, J. and Abbot, D.S., 2025. Can AI weather models predict out-of-distribution gray swan tropical cyclones?. Proceedings of the National Academy of Sciences, 122(21), p.e2420914122.
Lang, S., Alexe, M., Chantry, M., Dramsch, J., Pinault, F., Raoult, B., Clare, M.C., Lessig, C., Maier-Gerber, M., Magnusson, L. and Bouallègue, Z.B., 2024. AIFS–ECMWF’s data-driven forecasting system. arXiv preprint arXiv:2406.01465.
Dunn, R.J., Donat, M.G. and Alexander, L.V., 2022. Comparing extremes indices in recent observational and reanalysis products. Frontiers in Climate, 4, p.989505.
We strongly encourage anyone considering an application to contact us in advance for an informal chat about the project at an early stage. Please get in touch with Dr Tim Trent (University of Leicester) at [email protected] with any questions regarding this project. For further details on the Water and Climate Research Lab please visit www.wcrl.co.uk.
To apply to this project:
Applications must be submitted by 23:59 GMT on Wednesday 7th January 2026.