RainX Risk: GPU-Accelerated Modelling of Extreme Rainfall Risks in West Scotland - PhD

University of Glasgow

Glasgow, UK 🇬🇧

Project institution: University of Glasgow

Project supervisor(s): Dr Sebastian Mutz (University of Glasgow), Dr Daniela Castro-Camilo (University of Glasgow), Dr Dave McKay (University of Edinburgh), Dr Vinny Davies (University of Glasgow)

Research area: Geologic Hazard Analysis, Prediction and Digital Twinning

Extreme rainfall events in Scotland impact health, infrastructure, and economy through flooding, mass wasting, and water quality – some of the most urgent climate change related issues in the country [Sniffer, 2021]. However, the risk, i.e., the probability of harmful consequences or potential losses, is determined by the co-occurrence of extreme rainfall with other factors, such as the duration of the current rain period, tides, infrastructure, exposure, and more. 

This project will identify – in collaboration with local government (incl. the Glasgow City Council) – combinations of factors that increase the risks associated with extreme rainfall, with particular attention to compound extreme events. Building on advances in extreme value statistics and statistical downscaling, the project will develop GPU-optimised predictive models that are directly coupled to weather forecasts to estimate the probability of extreme rainfall and, ultimately, quantify the associated risks and uncertainties under different combinations of contributing factors. 

Methods: 

The primary goal of the project is the prediction of extreme rainfall related risks in West Scotland. This will be achieved through the following steps: 

The prediction of rainfall extremes will directly build on the lessons and developments of the ExaGEO project “RainX”, and use statistical downscaling [e.g., Hewitson et al., 2014; Boateng and Mutz, 2023] and advances in extreme value theory [Friederichs, 2010; de Carvalho et al., 2026, Chs. 9, 10, 24, 25]. Physics-informed neural networks incorporating extreme value theory will also be explored as a complementary or alternative approach to downscaling. This component will build on ongoing methodological developments within the supervisory team, providing an opportunity to investigate physics-informed downscaling while maintaining the project’s primary focus on rainfall-related risk prediction. 

Extreme rainfall will then be contextualised with other factors, such as tides, to assess the associated risks. The combinations of conditions associated with high risk will be characterised using thresholds and multivariate statistical models. Risk categories will be defined based on both the joint behaviour of these environmental variables and impact-relevant thresholds identified in collaboration with local government, allowing predictions to be classified into different levels of risk. 

The identification of compounding factors contributing to impacts on health, infrastructure, and economy will be conducted in collaboration with local government. This will ensure (1) the relevance of predicted impacts to risk management strategies and (2) that the project results can be integrated into existing decision-support models, including Glasgow City’s Environmental Digital Twin [Daravanis and McAllister, 2026]. 

The developed models will leverage Fortran’s native constructs for parallelisation, as well as increased GPU offloading support, to develop GPU-optimised models that predict extreme rainfall related risks based on multiple data steams (weather forecasts and compounding factors). 

Teaser Project 1 Objectives: Classification into Risk Categories 

The variable space associated with high risk will be described through multivariate probability density functions. Among others, a Bayesian classification approach [e.g., Tebaldi et al., 2005] and Random Forests classification [Breiman, 2001] will be used for the categorisation of specific predictions for extreme rainfall and other factors into different risk categories. The classification models are applied directly to the predictions of extreme rainfall and the features extracted from the different data streams to predict risk levels. 

Teaser Project 2 Objectives: Physics-informed Neural Networks for Environmental Extremes 

This project will explore physics-informed neural networks combined with extreme value theory for modelling environmental extremes. Building on ongoing research within the supervisory team, the student will implement and evaluate existing approaches using simulated or environmental data, focusing on their ability to capture tail behaviour, quantify uncertainty, and improve predictions of rare events. The project will provide experience in extreme value statistics and physics-informed machine learning, while helping identify opportunities for methodological improvements. 

69 days remaining

Apply by 31 January, 2027

POSITION TYPE

ORGANIZATION TYPE

EXPERIENCE-LEVEL

DEGREE REQUIRED

Register for World Toilet Summit 2026 - Thailand - November 17-19, 2026