PhD Studentship in Atmosphere, Oceans and Climate

Project title: Improving river streamflow forecasts using deep learning techniques

Department/School: Meteorology

Supervisors:  Kieran Hunt, Hannah Cloke, Florian Pappenberger, Christel Prudhomme

Project Overview:  Current-generation physics-based river discharge models require high-resolution data to be accurate. These constraints result in river discharge forecasts having a large computational cost and high sensitivity to errors in the underlying weather forecast: storms forecast in the wrong location or with the wrong intensity will produce hydrographs that differ wildly from reality. As with many such complicated non-linear processes, artificial neural networks (ANNs) have been applied to the problem with increasingly useful results.

Recent research has shown that a type of ANN known as a long short-term memory network (LSTM) is highly skilful at modelling river discharge when fed observational data, leading to interest in their potential use for forecasting. In this project, you will design a state-of-the-art LSTM to produce operational forecasts of river streamflow using data from ECMWF’s Integrated Forecast System (IFS) and Global Flood Awareness System (GloFAS). This will involve using sensitivity tests to determine the optimal network architecture and working with ECMWF scientists to incorporate IFS and GloFAS ensemble members and investigate potential avenues to reduce the impact of biases (persistent forecast errors) in the underlying IFS/GloFAS forecasts. This work will be initially carried out over Europe, but there will also be scope to test and refine the LSTM model in more data sparse areas, such as Africa and South Asia.

This is an exciting opportunity to be at the cutting edge of AI developments in global forecasting. Because LSTMs are relatively cheap to run once trained, you will have the opportunity to apply your work in a number of different domains. These might include tuning the LSTM to use output from climate models, leading to projections of discharge for selected rivers in future climate scenarios; reconstructing historical streamflow data where observations are unavailable; or, identifying emergent hydrological behaviours in catchments that current-generation physics-based models cannot reproduce, such as rain-on-snow events leading to high discharge.

Eligibility: 

• Applicants should hold or expect to gain a minimum of a 2:1 Bachelor Degree or equivalent in physics, maths, computer science, or a related discipline.
• Due to restrictions on the funding this studentship is open only to citizens of ECMWF member states. https://www.ecmwf.int/en/about/who-we-are/member-states

Funding Details:

• Starts 18 September 2023
• 3-year award plus a possible 6-month extension
• Funding covers full tuition fees plus UKRI stipend

How to apply:  

To apply click “Apply for a programme”, create your account and use the link sent by email to start the application process.   During the application process please select the PhD in Atmosphere, Oceans and Climate

*Important notes*

1. Please quote the reference ‘GS23-002’ in the ‘Scholarships applied for’ box which appears within the Funding Section of your on-line application.
2. When you are prompted by the online application system to upload a research proposal, please omit this step as the project is already defined.

Application Deadline:  20th January 2023

Further Enquiries:  Please contact Kieran Hunt (k.m.r.hunt@reading.ac.uk).

Please note that, where a candidate is successful in being awarded funding, this will be confirmed via a formal studentship award letter; this will be provided separately from any Offer of Admission and will be subject to standard checks for eligibility and other criteria.