Quantifying the likelihood of persistent extreme weather events in a changing climate
Background
Persistent extreme weather events such as heatwaves, cold snaps, or particularly rainy and/or stormy seasons have huge socio-economic impact; estimating the probabilities of their occurrence which can inform assessments of health or flood risks is therefore an important task. These probabilities are likely to change in a nonlinear way under climate change. A particular interest lies in events of unprecedented duration and/or severity. Mathematically, persistent extremes are described as extreme values of averages of quantities like temperature, precipitation or vorticity over certain extended time periods. Classical extreme value analysis is of limited use here due to the rarity of the events, the non-stationarity, and the resulting scarcity of the data. A further question is to link local persistent extremes to large- or synoptic-scale atmospheric circulation regimes which may help understand physical drivers and serve as mediators to facilitate prediction of the extreme episodes.
PhD project
This project will develop and explore novel data-driven techniques for quantifying the likelihood of persistent weather extremes. The approach is based on large-deviation theory, originally introduced in statistical physics to quantify probabilities of rare trajectories in dynamical systems, characterised by extreme values of averages of observables of interest along the trajectory. Averages both in time and in space can be considered. The central task is to estimate the so called large-deviation rate function, together with its uncertainty; once convergence is established the method allows for extrapolation to events of so far unobserved duration and severity. The rate function will be time-dependent in a changing climate and may be represented using machine learning techniques. Multivariate extensions of large-deviation theory are possible to quantify the likelihood of compound extreme events. We will apply the methodology to data sets from the large-scale CMIP climate model simulations and the convection-permitting UK Climate Projections (UKCP18) as well as observational data sets such as the Central England Temperature time series and ERA5 reanalysis. The project will also look at advanced interpretable machine learning-style clustering techniques targeted to optimally link local persistent extremes to large-scale atmospheric regimes. Additionally, the project will also apply these techniques to retrospectively investigate past extreme weather warnings issued by the National Severe Weather Warning Service (NSWWS).
The research will be carried out in close collaboration with two project partners, the Met Office and the company JBA Risk Management. The Met Office’s main interest here is in UK heatwaves and extreme precipitation, whereas JBA Risk Management is particularly interested in potential applications to high cumulative precipitation events and their implications for flood risk.
Applicant Profile
We are looking for students with a strong background in a relevant subject such as mathematics, statistics, physics, or meteorology who have a keen interest in applying these skills to problems in weather and climate science. Prior knowledge and experience in weather and climate science is an advantage but not mandatory.
Other Information
https://empslocal.ex.ac.uk/people/staff/fk206/
https://iopscience.iop.org/article/10.1088/1742-5468/ab02e8
https://link.springer.com/article/10.1007/s40766-021-00020-z
