Using AI to identify risks to infrastructure in UK high-resolution deformation data.

Summary

Built infrastructure in the UK is at risk from ground movements due to subsurface processes. Radar data from the Sentinel-1 constellation is being processed by University of Leeds spinout company, SatSense Ltd, which will produce high-resolution, high-accuracy, up-to-date motion products for the whole UK. This project will test and develop Artificial Intelligence (AI) methodologies to identify areas where infrastructure may be at risk from potentially catastrophic ground movement. If successful, the results could be of high value to the UK economy by providing information that can be used to mitigate risk.

Project Description

Satellite radar interferometry (InSAR) can monitor the motion of scatterers on the ground with high accuracy and high spatial resolution (Hooper et al. 2012). InSAR has been used for a wide variety of applications including monitoring deformation due to tectonics (Elliott, Walters and Wright 2016), volcanology (Pinel, Poland and Hooper 2014), glaciology (Joughin, Smith and Abdalati 2010), ground water flow (Bell et al. 2008), landslides (Hilley et al. 2004), resource extraction (Perski et al. 2009) and subsurface CO2 injection (Rutqvist, Vasco and Myer 2010). The launch of the Sentinel-1A/B satellites in 2014/16, as part of the EU Copernicus programme, has made radar imagery widely and freely available over the entire planet – all of Europe is imaged at least twice every 6 days. SatSense Ltd was launched in April 2018 as a spinout from the University of Leeds to develop commercial applications of InSAR using the latest algorithms developed by Leeds academics (e.g. Spaans and Hooper 2016). SatSense Ltd will be producing high-resolution deformation data over large regions in near-real time.

This wealth of InSAR data presents a huge opportunity to monitor infrastructure at risk of damage on a national scale. But the data volume is also a challenge – how can we efficiently, rapidly and automatically extract information of maximum benefit to end users. Furthermore, deformation occurs due to a variety of different causes that have different spatial and temporal patterns. To meet these challenges, machine learning algorithms have already shown some promise for volcanology (Ebmeier 2016). However, this approach is not able to classify the type of deformation, nor forecast how it will evolve. Deep learning approaches used in artificial intelligence (Varol, Laptev and Schmid 2017) have the potential to solve this problem and will be explored in this studentship.

The aim of this studentship is to use the latest developments in Artificial Intelligence to develop algorithms that can analyse large InSAR data sets automatically and identify areas where ground motion poses a risk to infrastructure. Specifically, the student will

1. Develop a suite of case studies using UK deformation data, identifying different areas where deformation has caused damage to infrastructure and characterising the temporal and spatial patterns of deformation.

2. Test and implement deep learning approaches, developing algorithms that can identify areas at risk from damaging ground movement.

3. Work with SatSense Ltd to implement algorithms for real-world applications.

The student will be supported by the supervisory team whose expertise covers all aspects of the project.

Student profile

The project would suit a numerate student with a background in computer science, mathematics, earth sciences or similar disciplines, who is enthusiastic about developing new approaches to working with large data sets. The student will be provided with training in state of the art methods. The student will have the opportunity for a formal placement with SatSense Ltd. They will also be part of the UK Natural Environmental Research Council’s Centre for the Observation and Modelling of Earthquakes, Volcanoes and Tectonics (COMET) and will have the opportunity to interact with COMET students with different skills and backgrounds from across the UK.