University of Edinburgh logo British Antarctic Survey logo National Oceanographic Centre logo University of Leeds logo


Image processing and machine learning applied to Earth Observation of the cryosphere

Image processing and machine learning applied to Earth Observation of the cryosphere

Supervisors: Robert Bingham, Noel Gourmelen (UoE); Hamish Pritchard, Anita Faul (BAS); Martin Ewart (EarthWave)

Background and Rationale

A number of recent studies have begun to highlight the potential to analyse remote sensing imagery to monitor changes that have occurred to the Earth’s ice sheets, ice caps and glaciers over the last few decades. A common technique has been to map changes to glacier frontal positions using optical imagery, which has shown significant and widespread retreat in locations such as the Antarctic Peninsula (Cook et al., 2016), parts of East Antarctica (Miles et al., 2013) and in parts of Greenland and glaciated Arctic islands (Carr et al., 2013). In West Antarctica, Christie et al. (2016; 2018) have mapped pervasive retreat of the grounding line over the last 40 years using a combination of Landsat and InSAR. These studies are just opening the door to the vastly wider potential for deploying remote sensing data and auxiliary information, where the ambition is to upscale activities from manual mapping of features such as ice fronts and grounding lines to more automated techniques inherited from machine learning. The ultimate goal of this project is to develop a framework for making best use of the existing remote sensing and geophysical data to map changes around the polar ice masses for the last 50 years, and to use these data to inform us on the processes responsible.

Aim and Key Research Questions

The aim of the proposal is to formalise a framework for making best use of available remote sensing (e.g. Landsat, MODIS, ASTER, SPOT) and geophysical data (radar sounding) to capture past change to polar ice masses. It is anticipated that the initial focus will be on Antarctica, but the ambition is to develop a workflow applicable to all polar ice masses. There is considerable flexibility in the direction that the project may ultimately take.

Key research questions are as follows:

  1. How can mapping ice front and grounding-line change be optimised taking advantage of techniques developed in image processing/analysis and/or machine learning?
  2. Over decadal timescales, are ice-front changes around Antarctica suitable proxies for grounding-line changes?
  3. Have changes at the ice front been transmitted inland in terms of optically-detectable signals of ice-flow change?
  4. What are the underlying causes of the glaciological observations recovered by these techniques?

Indicative timescale of activities and training

You will be supervised by glaciologists and image-processing experts based in Edinburgh (UoE and EarthWave) and the British Antarctic Survey, Cambridge. You will gain strong expertise in using a mix of techniques in remote sensing, image manipulation, glaciological interpretation, and ice-sheet modelling to build an archive of past ice change that will be of high impact to the scientific community. The project has some flexibility as regards its ultimate direction, but an indicative outline of some key stages is as follows:

  1. Apply techniques of Christie et al. (2016) to map changes to grounding-line position around unmapped West Antarctica, and then on to East Antarctica.
  2. Develop techniques to move towards automation of grounding-line and ice-front mapping.
  3. With Antarctic archive, explore relationship of mapped changes to e.g. El Nino Southern Oscillation.
  4. Explore transmission of mapped grounding-line/frontal changes inland through geometry of radar-imaged internal layers.
  5. Translate above methodologies to Greenland Ice Sheet and/or additional polar ice caps.

Required skills and qualifications prior to application

We seek an enthusiastic student with a suitable Undergraduate and/or Masters Degree qualification equipped with quantitative skills. Suitable backgrounds may be in Informatics, Maths, Physics, Earth Sciences or Physical Geography. Some prior experience of machine learning techniques is desirable.

References (further reading):

Bingham RG et al. (2015) J. Geophys. Res., 120, 655-70; Carr JR et al. (2013) J.Geophys. Res., 118, 1210-1226; Christie, FWD et al. (2016) Geophys. Res. Lett., 43, 5741-5749; Christie, FWD et al. (2018) Cryosphere, 12, 2461–2479; Cook AJ et al. (2016) Science, 353, 283-286; Miles, BWJ et al. (2013) Nature, 500, 563-566.