Understanding uncertainty in risk modelling of European windstorms: the East Atlantic pattern and history matching
Background
Extra-tropical cyclones are fascinating storms that have devastating impacts on society. For example, in winter 2013/14, more than 10 extreme storms passed over Europe, leading to total insured losses of more than $3.3 billion. Losses due to such events are expected to rise due to climate change. Insurers make decisions about such losses by using stochastic event simulations from a small set of catastrophe models. Such models have many simplifying assumptions and so give uncertain predictions of true losses. Furthermore, exact shifts in locations of highest risk are uncertain due to modulations in large-scale modes of variability, with the East Atlantic pattern being most important for the most exposed parts of Europe. There is uncertainty how this mode of variability might change due to climate change and the slowdown of the Atlantic deep ocean circulation.
PhD project
This thesis will address two challenges:
1. Local risk hotspots due to the East Atlantic Pattern
Recent work suggests that windstorms in W. Europe will strengthen in the future due to the N. Atlantic warming hole intensifying the winds near Europe (Priestley et al. 2026). This is closely related to a mode of variability known as the East Atlantic Pattern (EAP). This thesis will address the following questions:
a) How have extreme wind gust speeds depended on the EAP in previous winters?
b) How will the EAP change in the future (e.g. due to slowdown or collapse of the Atlantic Meridional Overturning Circulation) and what impact will this have on extreme storm risk?
c) How do future changes in EAP in climate models depend on their historical state?
This will involve fitting spatio-temporal extreme value models to windstorm footprints from datasets such as the 7th Climate Model Intercomparison Project (CMIP7). The approaches will improve understanding of how maximal winds might change in the future.
2. Uncertainty Quantification (UQ) in catastrophe models
How can loss simulations from a set of catastrophe models be adjusted (e.g. by importance resampling) to account for individual and common discrepancies in how such models represent extreme historical wind speeds? This is often done subjectively and using ad hoc approaches but could benefit from a more rigorous fundamental UQ underpinning. This component will involve the development and application of UQ methods (emulation, history matching), in particular:
a) Investigating uncertainties/assumptions in catastrophe models (vulnerability functions, uncertain inputs), and calibrate (history match) the unknown inputs based on observations;
b) How to combine predictions across multiple models, accounting for uncertainty/discrepancies between simulations and past observations, propagating uncertainty in inputs through to uncertainty in outputs, to produce better calibrated predictions of future losses.
Applicant Profile
This project will require the ability to develop advanced statistical methods and apply them to large meteorological data sets. An undergraduate degree in a quantitative subject such as statistics/mathematics/data science/computer science/physics/climate science is desirable and an interest in weather and climate science would be beneficial.
Other Information
https://www.wtwco.com/en-gb/insights/research-programs-and-collaborations/willis-research-network
https://experts.exeter.ac.uk/961-david-stephenson
Priestley, M.D.K., Stephenson, D.B., Scaife, A.A., and D. Bannister (2026) Environ. Res. Lett. https://doi.org/10.1088/1748-9326/ae845f (in press).
