October 12, 2022

Early warning of reservoir dam failure: harnessing machine learning and novel ground imaging for enhanced hazard assessment

Resistivity imaging of a large reservoir dam for condition assessment and leak detection

This is an exciting opportunity to explore deep learning (DL) methods, in combination with geophysical and hydro-mechanical models, to extract information on the stability of dams and other earthworks.

The collapse of the auxiliary spillway of the Toddbrook Dam following heavy rain during summer 2019 brought significant public attention to potential hazards from failure of large engineered earthen structures, In the UK alone it is estimated that there are >3000 embankment dams, with an average age of >100 years old. Unlike traditional surface or point observations, modern geophysical imaging can dynamically sense deep into the interior of earthworks.

Promising developments in deep learning on feature extraction, joint interpretation, and fast inference have opened doors to provide rapid early warning from high-dimensional geophysical data. You will contribute towards geophysical early warning and creating “digital twins” of dams by developing and training deep learning models that simulates the geotechnical response and the corresponding geophysical signal under different scenario. Key to these DL models is their ability to make fast predictions and inference, embed the complex input/output relationships driven by physical processes, and extensible to include other data types. You will explore the response (with uncertainty estimates) of the dam under various weather (e.g. UKCP18 scenarios) and geotechnical states and apply the methods to UK dams and large-scale tank experiments.

What’s in it for you

You will gain highly marketable and transferable skills in data science and machine learning, and in geophysical methods, inverse methods and uncertainty analysis. You will develop links with external organizations such as BGS and our industrial and international partners. You will join vibrant research communities, e.g. Centre of Excellence in Environmental Data Science (, LU-UKCEH). The project is supported by industrial partner HR Wallingford, giving you an opportunity to apply your research in an industrial setting.

We are seeking applications from graduates or those who expect to graduate in 2023 with a good BSc (2:1 level or equivalent) or Master’s degree. You should have a strong background in Data Science, Computer science, Maths and Statistics, Earth and Environmental Sciences, Geophysics, Physics or Engineering. Experience of coding (e.g. python, R, MATLAB, Julia) and experience with machine learning is desirable. You must have demonstrable potential for creative, high-quality PhD research, broad interest in geo-environmental problems, and a track record of making sense of data.

Please contact Dr. Michael Tso ( or Prof. Andrew Binley (