About the Project
Persistent droughts represent one of the most consequential climate extremes, exerting severe impacts on water resources, agricultural productivity, ecosystem integrity, energy generation, and public health. Quantifying and managing these risks requires an understanding of the plausible severity of worst-case drought scenarios under present and future climate conditions. However, climate change introduces additional complexity: while UK projections indicate warmer, drier summers and wetter winters, the implications for multi-year droughts remain highly uncertain. This uncertainty is critical because prolonged droughts, though rare, can have disproportionate societal and environmental consequences. They compromise water availability for households and industry, disrupt agricultural productivity, threaten biodiversity, reduce energy generation from hydropower, and pose serious risks to public health.
Assessing events of such low probability demands climate time series spanning thousands of years to estimate their likelihood and characteristics. Neither observational records nor computationally expensive high-resolution climate models can provide datasets of this magnitude. This limitation necessitates innovative approaches to efficiently explore the tails of climate distributions.
Rare Event Simulation (RES) offers a powerful framework to overcome these constraints (Del Moral, 2005; Wouters, 2023). RES accelerates the sampling of rare extremes by running large ensembles of climate model simulations in parallel. At intermediate stages, simulations unlikely to produce an extreme event are terminated, while those exhibiting early indicators of such events are cloned. This adaptive resampling strategy concentrates computational resources where they are most informative, enabling robust estimation of probabilities for events that would otherwise be computationally prohibitive.
A central methodological challenge is determining which ensemble members are “promising” and should be cloned. This project addresses this challenge by developing machine learning techniques to extract predictive features from evolving climate states and forecast the likelihood of drought development (Mascolo, 2025). Building on these predictive models, we will design new algorithms to optimally integrate machine learning within the RES framework, ensuring efficient allocation of computational resources.
The project will deliver novel machine learning methods for early detection of developing droughts in climate simulations, together with a new RES workflow that leverages predictive modeling for.
optimal ensemble resampling. It provides the successful applicant with a unique combination of expertise in advanced machine learning, probabilistic computational techniques, and climate dynamics.
The outcomes will have broad applicability beyond droughts, supporting the growing international effort to apply RES to a range of climate extremes, including heatwaves and intense rainfall. By enabling RES integration with state-of-the-art high-resolution climate models, this research will advance the capability of climate science to quantify and manage risks associated with rare but high-impact events.
Funding Notes
A full UKRI stipend plus home-level PhD tuition fees, details are found below:
References
J. Wouters, R. Schiemann and L. Shaffrey, “Rare event simulation of extreme European winter rainfall in an intermediate complexity climate model”, Journal of Advances in Modeling Earth Systems, 15 (2023) https://doi.org/10.1029/2022MS003537 1942-2466.
Del Moral, P., Garnier, J., 2005. Genealogical particle analysis of rare events. The Annals of Applied Probability 15, 2496–2534. https://doi.org/10.1214/105051605000000566
Mascolo, V., Lovo, A., Herbert, C., Bouchet, F., 2025. Gaussian Framework and Optimal Projection of Weather Fields for Prediction of Extreme Events. Journal of Advances in Modeling Earth Systems 17, e2024MS004487. https://doi.org/10.1029/2024MS004487
