Projects that started in 2025 (UNRISK cohort 1)
Physics-informed machine learning for decision making related to future extreme weather events
- “MC” Benoist
- Mathematics and Statistics, University of Exeter
- Supervisors: Stefan Siegert and Jemma Shipton
- Partners: KAUST
This project develops physics-informed machine learning (ML) methods to improve predictions of extreme weather events under climate change. Probabilistic ML models can estimate the risk of short-lived, localised events like severe precipitation, but ensuring physical realism remains a challenge. This research will explore how to embed physical constraints into ML models—through architecture design, loss functions, or augmented training data—and evaluate their impact on predictive skill. It will also advance techniques for generating spatiotemporal probabilistic forecasts and robust evaluation metrics. The goal is to deliver accurate, physically consistent predictions to support climate risk assessment and resilience planning.
Novel statistical AI approaches for modelling and evaluating extreme windstorm risk
- Serin Mary Binoy
- Mathematics and Statistics, University of Exeter
- Supervisors: David Stephenson, Matthew Priestley and Theo Economou
- Partners: Willis Towers Watson
My project investigates how advanced statistical AI methods can improve our understanding and modelling of extreme windstorm risk. Extra-tropical cyclones cause significant societal and financial impacts, and current catastrophe models used by insurers rely on simplifying assumptions. This project will apply modern techniques such as Bayesian hierarchical spatial models to analyse windstorm footprint data across Europe, Japan, and beyond. By characterising extreme wind speeds and their relationship to climate variability and change, the research will support more accurate risk assessments and help insurers develop better protection strategies. Collaboration with Willis Towers Watson and the Met Office ensures real-world relevance.
Advanced uncertainty quantification for decision support: Balancing risks and preferences for policy decisions
- Rose Crocker
- LEEP, University of Exeter
- Supervisors: Amy Binner and Danny Williamson
My project aims to develop advanced tools for decision-making under uncertainty, focusing on land-use strategies for achieving goals like the UK’s Net Zero target. Policymakers face complex choices influenced by climate, economic, and regulatory uncertainties. This project combines environmental and economic modelling with cutting-edge uncertainty quantification and preference elicitation techniques. By creating transparent, user-friendly tools, it will allow decision-makers to explore alternative strategies, assess risks, and incorporate their priorities. Methods such as emulation and model calibration will enable rapid, evidence-based evaluation of options, supporting adaptive, robust policies for sustainable land management and long-term societal and environmental outcomes.
Quantifying the role of land cover in ‘slowing the flow’ for flood risk reduction in a changing climate
- Jake Senior
- School of Geography, University of Leeds
- Supervisors: Steph Bond and Joe Holden
- Partners: National Trust
My project investigates how land cover management can reduce flood risk through Nature-Based Solutions (NBS) in a changing climate. While NBS such as “slowing the flow” are widely promoted, their effectiveness under future climate uncertainty remains unclear. This research will combine field data, hydrological modelling, and data science to quantify how vegetation and surface roughness influence overland flow and flood risk. It will also explore how climate-driven changes in storms and vegetation affect model predictions and uncertainty. The findings will inform practical strategies for flood risk reduction and support evidence-based policy for sustainable catchment management.
Predicting biodiversity loss in mountain rivers due to glacier retreat
- Arthur Lamoliere
- School of Geography, University of Leeds
- Supervisors: Lee Brown and Jonathan Carrivick
My project aims to predict biodiversity loss in glacier-fed mountain rivers as glaciers retreat under climate change. While physical processes in these rivers are well understood, major uncertainties remain about species responses and extinction risks, limiting effective conservation. This project will reduce these uncertainties by improving species distribution models (SDMs) through better datasets, advanced modelling, and integration of climate and hydrological projections. Work will include analysing data gaps, collecting new biological data in regions such as the Alps or Himalayas, and evaluating uncertainty from multiple modelling approaches. The goal is to inform conservation strategies for vulnerable mountain ecosystems.
Reducing uncertainty in the effect of clouds on climate change
- Soffie Wisniewska
- School of Earth and Environment, University of Leeds
- Supervisors: Ken Carslaw and Paul Field
- Partners: Met Office
My project focuses on reducing one of the largest uncertainties in climate projections: the role of clouds. Cloud feedbacks and aerosol–cloud interactions strongly influence global warming, yet current climate models struggle to represent these processes accurately. This research will combine high-resolution weather-scale simulations, satellite and aircraft observations, and advanced data science techniques to identify the physical processes driving uncertainty. By using machine learning emulators trained on large ensembles of model simulations, the project will explore ways to constrain cloud behaviour, improve model realism, and deliver more reliable long-term climate predictions.
Using new high-resolution ensembles to quantify uncertainty in projections of African climate processes
- Ella Thomas
- School of Earth and Environment, University of Leeds
- Supervisors: John Marsham and Ben Maybee
- Partners: Met Office
My project explores the connections between large-scale drivers and convective precipitation in West Africa. It uses a new ensemble of convection-permitting simulations run over continental Africa for two 10-year periods (current and future climate). Rainfall in West Africa is dominated by convection, which acts on a small spatial scale but produces large amounts of energy, feeding back into large-scale circulations. Convective-scale simulations are necessary to resolve these processes and develop better physical understanding of the mechanisms controlling them. This understanding will be used to reduce uncertainty in user-relevant metrics such as monsoon onset and future projections of rainfall extremes.
Using new high-resolution ensembles to quantify uncertainty in projections of African climate processes
- Dylan Dearnaley
- School of Earth and Environment, University of Leeds
- Supervisors: Cathryn Birch, Fleur Loveridge, Susanne Lorenz
- Partners: Met Office, rail industry
My research addresses the growing risk of landslides to UK rail infrastructure under climate extremes. Landslides, often triggered by intense rainfall, already cause frequent disruptions and are projected to increase with climate change. The project will quantify the rainfall characteristics that lead to slope failures, assess how these may change in the future, and develop methods to combine and communicate uncertainties in rare event prediction. Working with rail industry partners and the Met Office, the research will create tools and best practices for risk management, early warning, and long-term adaptation strategies to protect critical transport networks.
The influence of physical process representations on regional and global-scale climate model output
- Celine Tchaghlassian
- School of Earth and Environment, University of Leeds
- Supervisors: Leighton Regayre and Ken Carslaw
- Partners: CICERO, Norway
My project examines how representations of physical processes in climate models influence global climate projection uncertainty. Aerosol–cloud interactions remain one of the largest sources of forcing uncertainty, masking the true extent of greenhouse gas warming. Using large ensembles of simulations with the UK Earth System Model (UKESM2) alongside multi-model perturbed parameter ensembles, the research will identify how cloud microphysics, aerosol and atmospheric dynamics contribute to uncertainty. Novel methods will disentangle structural from parametric causes of aerosol–cloud forcing uncertainty, providing insights into causes of model divergence, and guiding model developments that target improvements in climate prediction.
Leveraging machine learning algorithms for improved Arctic sea-ice prediction using the Met Office suite of models
- Qien Cai
- Earth Sciences, University College London
- Supervisors: Michel Tsamados
- Partners: Met Office
My project focuses on improving sea-ice forecasting by combining satellite observations, machine learning, and physical models. Satellite altimetry has transformed our ability to estimate sea-ice thickness and ocean circulation, and recent advances now allow thickness mapping even in summer months. This project will build on these breakthroughs by developing deep learning algorithms trained on long ocean–sea ice reanalysis datasets and high-resolution Arctic models. The aim is to enhance short-range and seasonal forecasts of sea-ice concentration and thickness, and compare machine learning predictions with traditional dynamical models, supporting better climate prediction and operational decision-making in polar regions.
Reducing uncertainty in climate risk perception to enhance resilience: A data science approach using the World Risk Poll
- Beth Dunstan
- Risk and Disaster Reduction, University College London
- Supervisors: Mohammad Shamsudduha and Carina Fearnley
- Partners: Lloyd’s Register Foundation
My project investigates how perceptions of climate risk evolve globally and how they align with actual climate hazards. Using the Lloyd’s Register Foundation World Risk Poll (2019–2023), it will analyse spatiotemporal trends in climate risk perception before and after the COVID-19 pandemic, focusing on differences across regions and socio-economic contexts. Advanced data science and visualisation methods will uncover drivers of perception gaps and their implications for resilience. By comparing perceptions with real hazard exposure, the research will inform strategies for effective risk communication and adaptation, with a particular emphasis on vulnerable nations in the Global South.
End-to-end machine learning quantification of hydrological uncertainties: from climate to flood risk management
- Valentin Brekke
- Statistical Science, University College London
- Supervisors: Serge Guillas and Erica Thompson
- Partners: RIKEN Center for Computational Science
My project aims to use machine learning to improve how uncertainties are quantified across the entire chain from climate projections to flood risk. Recent advances in deep learning allow the creation of fast, accurate emulators for complex hydrological processes, reducing computational costs while maintaining fidelity. Working with high-resolution models developed at RIKEN-CCS in Japan, this project will integrate climate, hydrology, and decision-making under uncertainty. The goal is to develop efficient ML techniques that propagate uncertainties through the modelling chain and support practical decisions for flood mitigation, water resource management, and infrastructure planning in a changing climate.
Inclusive storylines for sustainable governance of urban adaptation to uncertainties in future heat extremes
- Xumeng (Andy) Deng
- STEaPP, University College London
- Supervisors: Bipashyee Ghosh and Erica Thompson
My project explores how inclusive climate storylines can support urban adaptation to extreme heat under uncertainty. While heat extremes are a certain impact of climate change, uncertainties remain in their severity and in social and technological responses. Working with detailed case studies, the project will analyse climate projections alongside urban planning data to understand how heat events affect infrastructure systems and governance decisions. It will then develop evidence-based, plural storylines to engage diverse stakeholders and guide robust, inclusive policies. By linking physical climate science with governance challenges, the project aims to improve anticipatory planning for resilient urban futures.
Combining models and uncertainties to support flood risk assessment and mitigation strategies
- Conor Lamb
- STEaPP, University College London
- Supervisors: Erica Thompson and Arthur Peterson
- Partners: FloodRe
My project aims to explore, and improve, how models support decision making within flood risk, public policy and (re)insurance. The project partners with Flood Re, a UK government scheme aiming to ensure all UK residential properties can obtain affordable flood insurance. The scheme uses multiple levers, across public policy and (re)insurance, to fulfil its purpose. My project will use a variety of qualitative and quantitative methods to explore how models support decision making across Flood Re, in the context of uncertainty, and make actionable suggestions about how this process may be improved.
