Local recruitment: Deep learning and prediction of compound extreme floods in coastal areas for a sustainable future - PhD (U.K. nationals)

Title: Deep learning and prediction of compound extreme floods in coastal areas for a sustainable future

Lead supervisor: Dr Yunqing Xuan, Faculty of Science and Engineering, Swansea University

Email: y.xuan@swansea.ac.uk

Co-supervisor: Dr Xiangbo Feng, National Centre for Atmospheric Science; Dr William Bennett, Civil Engineering, Swansea University

Project summary

Climate change is intensifying extreme weather events, particularly in coastal regions where heavy rainfall, river discharge, storm surges and sea-level rise can combine to cause devastating compound floods[1][2]. These events threaten lives, infrastructure and ecosystems, yet remain difficult to predict.

This PhD project will develop advanced artificial intelligence (AI) and physics-informed models to better understand, simulate and predict compound floods across both short-term (hours to days) and long-term (years to decades) timescales. The research will explore how deep learning, especially Transformer-based neural networks[5][6] and Fourier Neural Operators (FNOs)[7][8], can capture the coupling between atmosphere, river and ocean processes. The goal is to achieve real-time forecasting capability and improve long-term risk assessment for sustainable coastal futures.

Scientific motivation

Traditional flood-prediction methods either rely on statistical models assuming fixed relationships among flood drivers or on physics-based models that are too computationally expensive for real-time use[3][4]. In reality, the links between heavy rainfall, storm surge and coastal inundation evolve under climate change.

This project bridges that gap by combining deep learning with physical understanding. AI models will be trained using real-world observations and high-resolution hydrological, meteorological and coastal simulations. The approach will reproduce both statistical dependencies and physical mechanisms of flood propagation, creating efficient and transferable models for forecasting and climate-adaptation planning.

Research objectives (Fig. 1)

1. Characterise dynamic interactions among flood drivers.

Using attention-based deep learning architectures such as Transformers, the student will analyse long-term datasets to reveal how rainfall, runoff, sea level and storm surge interact across space and time.

2. Develop fast, physics-informed flood simulators.

A Fourier Neural Operator (FNO) framework will emulate the behaviour of complex hydrodynamic models, efficiently representing flood dynamics and enabling near real-time simulations.

3. Apply and validate models in coastal case studies.

The student will apply the models to flood-prone regions such as the UK, Vietnam and Indonesia. Results will inform early-warning systems and long-term resilience planning for sustainable coastal management.

Supervision and synergy

The project benefits from a supervisory team with complementary expertise in hydrometeorological modelling, climate and weather extremes, and coastal engineering. Together, they provide a research environment covering the full pathway from atmospheric drivers of storms to their impacts on coastal flooding.

The lead supervisor offers expertise in hydrology, flood forecasting and AI-based environmental modelling. Collaboration with an atmospheric-science group at a national research centre adds strength in tropical cyclones, ocean–atmosphere interactions and storm-surge prediction. The coastal-engineering supervision contributes practical knowledge of estuarine processes, flooding mechanisms and nature-based resilience.

This synergy links meteorological extremes over the ocean with compound flooding along the coast. The team’s experience will guide the student in developing innovative, physically consistent AI models that are scientifically robust and operationally relevant.

Research environment and collaboration

The PhD will be based at Swansea University’s Zienkiewicz Institute for Modelling, Data and AI, in collaboration with the University of Reading’s National Centre for Atmospheric Science (NCAS). The student will benefit from international partnerships and datasets through Reading’s SEA-COAST [9] and FORWARDS projects, which study tropical storms, sea-level extremes and coastal flooding across Southeast Asia.Reading also has strong links with the European Centre for Medium-Range Weather Forecasts (ECMWF). Through initiatives such as GloFAS and the AFESP partnership, Reading researchers collaborate with ECMWF scientists on advancing Earth-system prediction and compound-hazard forecasting. This connection provides access to operational-grade data and opportunities to refine AI models within a world-leading forecasting environment.Training will include deep learning, hydrological and coastal modelling, high-performance computing and data assimilation. The student will gain transferable skills relevant to academia, government and industry, including AI-based forecasting, risk assessment and climate adaptation.

Expected outcomes and impact

The project will produce:

• AI-driven, physics-aware models for predicting compound floods across multiple timescales.
• New insights into changing relationships between rainfall, storm surge and river discharge in a warming climate.
• Open-access datasets and modelling toolkits for researchers and practitioners, supporting transparency and collaboration.
• Practical tools for coastal resilience, sustainable development and disaster-risk reduction.

This PhD provides a unique opportunity to work at the forefront of environmental data science, blending AI innovation with practical impact. The student will enhance scientific understanding and develop solutions to safeguard coastal communities against the increasing risk of compound flooding in a changing environment.

Training opportunities:

The student will receive comprehensive training in deep learning, hydrological and coastal modelling, and high-performance computing. They will have opportunities to work with the Met Office and the National Centre for Atmospheric Science through the SEA-COAST collaboration at Reading, gaining experience with operational storm and flood forecasting systems. Fieldwork and data collection in selected coastal regions such as Vietnam and Indonesia may be arranged through international partnerships. The student will also benefit from interdisciplinary workshops, research exchanges, and professional-development training within Swansea University’s Zienkiewicz Institute.

Student profile:

This project would be suitable for students with a Master’s degree in hydrology, meteorology, civil engineering, or a closely related environmental or physical science. A strong background in quantitative analysis, numerical modelling, or data science is essential. Experience with programming tools such as Python or MATLAB is highly desirable. The ideal candidate will have a strong interest in applying advanced computational and AI methods to real-world environmental challenges, and be motivated to work across disciplines linking weather, hydrology, and coastal flood risk under a changing climate. UKRI funding only covers Home fees which increase annually. For this project, the difference in the Home fees and international fees will be covered and international students will not be required to meet the difference themselves. 

Co-Sponsorship details:

This project includes a 3-month placement opportunity at IHE Delft.

Reference

1. Wang, H., Xuan, Y., Tran, T., Couasnon, A., Scussolini, P., Luu, L., Nguyen, H., & Reeve, D. (2023). Changes in seasonal compound floods in Vietnam revealed by a time-varying dependence structure of extreme rainfall and high surge. Coastal Engineering, 183, 104330, https://doi.org/10.1016/j.coastaleng.2023.104330
2. Bennett, W.G., Karunarathna, H., Xuan, Y., Badri Kusuma, M.S., Farid, M., Kuntoro, A.A., Rahayu, H., Kombaitan, B., Sepiaditi, D., Adi Kesuma, T.N., Haigh, R. and Amaratunge, D. (2023). Modelling compound flooding: A case study from Jakarta, Indonesia, Natural Hazards, http://doi.org/10.1007/s11069-023-06001-1
3. Wang, H. & Xuan, Y. (2023). Deep Learning of Extreme Rainfall Patterns Using Enhanced Spatial Random Sampling With Pattern Recognition. Special Publications (pp. 343-356). Wiley, https://doi.org/10.1002/9781119639268.ch12
4. Wang, H. & Xuan, Y. (2024). A Nonstationary Multivariate Framework for Modelling Compound Flooding. In Springer Water (pp. 407-428). Springer Nature Singapore. https://doi.org/10.1007/978-981-97-4072-7_26
5. Jiang, H., Lu, Y., Zhang, D., Shi, Y., & Wang, J. (2024). Deep learning-based fusion networks with high-order attention mechanism for 3D object detection in autonomous driving scenarios. Applied Soft Computing, 152, 111253.
6. Lin, G., Shen, C., & Lin, A. (2024). Higher-order Cross-structural Embedding Model for Time Series Analysis. arXiv preprint arXiv:2410.22984.Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A., & Anandkumar, A. (2020). Fourier neural operator for parametric partial differential equations. arXiv preprint arXiv:2010.08895.
7. Qin, S., Lyu, F., Peng, W., Geng, D., Wang, J., Tang, X., Leroyer, S., Gao, N., Liu, X., & Wang, L. L. (2024). Toward a Better Understanding of Fourier Neural Operators from a Spectral Perspective. ArXiv. https://arxiv.org/abs/2404.07200
8. SEA-COAST: Understanding and Prediction of Compound Ocean-Atmosphere Storms in the Tropics, the Met Office WCSSP Southeast Asia project, 2025–2028