Using Statistical Emulators to Quantify and Understand Urban Climate Risks

Background

As climate change intensifies, cities face growing risks from extreme weather events such as heatwaves and floods. Addressing these risks requires simulation models to forecast various scenarios. However, these models are often computationally intensive, which limits their use in analyses such as uncertainty quantification, where many simulation runs are needed. To overcome this issue, statistical emulators are used to approximate complex models with far lower computational costs. Emulators allow for faster analysis and a deeper understanding of uncertainty and climate risks.

Despite recent advances, important challenges remain: for example, how can we ensure the accuracy and interpretability of emulators when they are used to assess climate risks such as flooding? Advancing research in this area is critical to building more resilient urban systems and offers a valuable opportunity for STEM graduates to contribute to environmental solutions through innovative, data-driven methods.

PhD opportunity

Climate change is a critical global challenge, demanding urgent policy responses to manage risks such as flooding and extreme heat. High-fidelity simulation models are essential for studying the impacts of different climate scenarios. However, their substantial computational cost restricts their use in tasks such as scenario exploration and rigorous uncertainty quantification (UQ), both of which require many simulation runs.

This PhD project aims to address these challenges by developing advanced data-driven methods to quantify and interpret uncertainties in multi-fidelity climate simulators. The focus will be on statistical emulators, e.g. Gaussian processes, combined with adaptive design of experiments that automatically select new simulation runs to maximise information gain across different fidelity levels. This approach will enable efficient approximations of high-fidelity models and help overcome computational barriers in exploring climate scenarios and quantifying uncertainty.

Working at the intersection of statistics, machine learning, and climate science, the student will develop emulators to model urban impacts of climate extremes. The project will also investigate the limitations of emulators, including high dimensionality and extrapolation issues. Key research questions include how different heatwave and flood scenarios influence infrastructure resilience and population vulnerability in cities. To address these, the PhD will advance UQ methods and explainable AI tools tailored to support decision-making.

The PhD will offer interdisciplinary supervision and training, including collaboration with Dr Saves’ team at Université Toulouse Capitole in France. Their expertise in urban mobility, flash flooding, spatial planning, and climate risk modelling will strengthen the scientific underpinnings of the project and enhance its practical relevance across European urban environments. The project will also be co-supervised by Professor Jennifer Catto at Exeter.

Applicant Profile

We welcome applications from STEM graduates, particularly those with backgrounds in statistics, computer science, modelling, and optimisation, who are eager to address real-world challenges in climate change, sustainability, and urban decision-making. Prior exposure to interdisciplinary fields such as social sciences, urban modelling, or ecology is advantageous but not required. This PhD is ideal for candidates who are curious, highly motivated, and keen to bridge computational methods with pressing societal and environmental issues.

Key skills and interests:
• Proficiency in scientific computing and programming (Python, R or equivalent)
• Foundations in optimisation, statistical learning and machine learning
• A strong interest in environmental engineering, agent‐based systems and interdisciplinary science

Other information

• E. Cueille, D. Bodini, B. Gaudou, D. Grancher, P. Nicolle, O. Payrastre, M. Prédhumeau, I. Ruin, G. Terti, and N. Verstaevel. “”Assessing the impact of crisis cell decisions during flash flood””, In : International Workshop on Multi-Agent-Based Simulation, 2025
• H. Mohammadi and P. Challenor, “”Sequential adaptive design for emulating costly computer codes””, Journal of Statistical Computation and Simulation, 2025
• Manon Prédhumeau, “”Sustainable urban digital twins: Reducing, reusing, recycling models””, In: The 19th International Conference on Computational Urban Planning and Urban Management, 2025.