Tim Wright (Leeds), Andy Hooper (Leeds), Stuart King (Edinburgh), Sarah Douglas (Satsense Ltd)
Earthquakes strike suddenly and without warning and can impact a large geographic region. Rapid identification of the amount and distribution of damage is critical for local and international responders (1, 2). This PhD project will develop tools for rapid damage mapping following large earthquakes, exploiting recent advances in synthetic aperture radar imaging and machine learning.
Satellite observations can also be used to directly assess damage resulting from earthquakes and secondary hazards such as landslides (3). Methods that use high-resolution optical images are powerful but fail when there is cloud cover, and currently require significant human involvement to produce results; results from optical methods are typically not available on the timescales required for emergency response (4). Satellite radar (SAR) imagery can see through clouds and can be acquired night and day. The recent proliferation of SAR missions (3) means we will soon be able to obtain new SAR imagery for most earthquakes in less than 24 hours (5).
One approach that has been used to create damage-proxy maps automatically is to track changes in interferometric coherence in SAR interferograms (5, 6). These methods track changes in the scattering properties; they are powerful but have relatively low spatial resolution as the coherence is calculated from patches of pixels (Fig. 1). Spaans and Hooper (7) proposed a method (RapidSAR) that produces coherence images at the highest resolution of the satellite using stacks of pre-processed radar imagery. RapidSAR coherence maps are significantly sharper than coherence maps from conventional imagery, showing key infrastructure more clearly. However, calculating coherence requires pre-processing large stacks of radar imagery.
In COMET, we have developed a system for processing large quantities of Sentinel-1 satellite radar imagery over tectonic and volcanic areas (8). The student will exploit this to produce RapidSAR coherence, phase and amplitude time series from Sentinel-1 SAR data. They will test the use of various machine learning and deep learning classifiers (e.g., boosted decision trees, convolutional neural networks) to identify areas where rapid changes have occurred during an earthquake. This may involve use of individual observations or series of observations to construct a suitable feature set for change detection. The first step, which will involve the most research and experimentation, will be to test and develop algorithms using data from an earthquake with known distribution of damage, for example recent earthquakes in Nepal, New Zealand, and Italy. The student will then apply the method systematically to a suite of earthquakes that occur during the timeframe of the project, with the ultimate aim of implementing a routine rapid damage assessment system.
The project would suit a numerate scientist from a wide range of disciplines, including physical sciences, mathematics, and computer science. The student will receive a wide array of training in Earth Observation and Machine Learning as part of the CDT and will benefit from membership of the COMET network (http://comet.nerc.ac.uk).
This PhD is part of the NERC and UK Space Agency funded Centre for Doctoral Training "SENSE": the Centre for Satellite Data in Environmental Science. SENSE will train 50 PhD students to tackle cross-disciplinary environmental problems by applying the latest data science techniques to satellite data. All our students will receive extensive training on satellite data and AI/Machine Learning and field training. All students will experience extensive training on professional skills, including spending 3 months on an industry placement. See http://www.eo-cdt.org
Application support
More information for applicants can be found on our available Projects, How to Apply and dedicated FAQ webpages. You will also find additional resources such as How can a PhD help me with my career?