PhD Studentship: Deep Learning for Surface Water Flood Risk Mapping and Forecasting

Rationale

Flooding – the most wide-spread natural hazard – affects every country and region of the world. Flood risk is expected to increase due to climate change, as evidenced by recent recurring UK summer and winter floods. The UK climate projections (UKCP18) suggest a >10% increase in heavy rainfall by 2050, with much of this “very likely” to fall in a short period of time [1], causing more severe surface water flooding. This type of flooding threatens more UK people and properties than any other; 3.2 million properties in England alone. Reliable forecasting and early warning can improve preparations, response and recovery, but rapid onset and localised extent make observing and predicting surface water flooding from intense rainfall technically challenging, and our ability to provide reliable, detailed forecasts remains limited [2]. We recently made a significant contribution by developing a new high-performance hydrodynamic system to forecast surface water flooding across an entire catchment at unprecedented resolution [3].

But the latest developments in AI and data analytics technologies have not yet sufficiently exploited to advance operational surface water flood forecasting; uncertainties in different components a forecasting system, e.g. numerical weather predictions and flood dynamics modelling, need to be better understood, quantified and minimised.

Methodology

The aim of this exciting PhD project is to harness the latest developments in high-performance computing and deep learning (DL) technologies to address some of the key technical challenges, and finally demonstrate a DL-enabled system for mapping, risk assessment and real-time forecasting of surface water flooding from intense rainfall. The project will deliver the following key research tasks:

• Develop physis-informed DL models to integrate rainfall observations from different sources and identify conditions associated with very extreme rainfall from the outputs of convection-permitting numerical weather models, and then emulate such numerical weather predictions to create reliable real time forecasts or large ensembles.
• Integrate the improved weather forecasts with the Loughborough in-house High-Performance Integrated hydrodynamic Modelling System (HiPIMS) to forecast in real time the surface water flooding process at a meter-level resolution for assessing flood impact/risk on individual buildings/objects. HiPIMS will be ML-enabled to support rapid ensemble forecasting.
• Design and perform systematic numerical experiments to better understand and quantify the uncertainties in different steps, their interaction and propagation through the entire flood forecasting procedure, and interpret their implication on the final flood forecasting and risk products.
• Demonstrate the system for real-time flood mapping, risk assessment and probability forecasting in a selected case study site.

This project is part of the NERC funded Flood-CDT studentship competition. For more information, please visit the Flood-CDT website.

Please see this PhD project’s dedicated webpage for more information via the above ‘Apply’ button.

Additional Funding Information

Studentship type – UKRI through Flood-CDT (flood-cdt.ac.uk). The studentship is for 3.5 years and provides a tax-free stipend of £18,622 per annum plus tuition fees at the UK rate. Excellent International candidates are eligible for a full international fee waiver however due to UKRI funding rules, no more than 30% of the studentships funded by this grant can be awarded to International candidates.

For details on what documents you will be required to submit for this, please see the dedicated project page, available via the Apply button.

Advert information