skip to content

Cambridge NERC Doctoral Training Partnerships

Graduate Research Opportunities
Brief summary: 
Developing AI methods to count whales in very high resolution satellite imagery.
Importance of the area of research concerned: 
Many species of whale are facing increased anthropogenic threats. One of the main challenges for mitigating threats is poor understanding of whale distribution and abundance. Although large, whales are cryptic; spending limited time at the surface and ranging over large oceanic areas. Traditional survey techniques have limited range and are expensive to conduct over large areas. One solution is to use very high-resolution satellite images, which can identify surfacing whales, which, along with known dive profiles, can be used to estimate abundance. Several recent papers have shown the potential of this approach, but also highlighted the problem of manually counting whales within imagery, currently one of the greatest barriers for development of the technique. Detailed manual counting is inefficient, error-prone and variable. This PhD will develop Deep Learning tools to automatically identify whales in satellite imagery. This is a key step towards measuring whale densities over large geographic scales and assessing distributions in key areas where there are environmental risks and potential for conflicts with human activities.
Project summary : 
This PhD application builds on previous BAS/UCAM work including two MSc projects and three on-going collaborations on the topic of automating whale finding using CNNs. British Antarctic Survey has the world’s largest collection of satellite/whale ground truthing data and the most experienced team in this research space. The student will draw on the experience of these projects and parallel BAS work on seals and albatross to build, refine and implement better solutions for the automation of whale finding in satellite imagery. This can include identifying multiple whale species in multiple habitats, use of UAV imagery to augment test datasets, probabilistic assignment of whales to species, and development of an auto-ID pipeline for whale detection.
What will the student do?: 
Potential themes of the PhD include: • Perfecting a single species Convoluted Neural Network whale identifier for a number of species: gray, humpback, beluga and rorqual whales • Improvement of algorithms by adding confounding factors: waves, boats, rocks and clouds and testing multiple CNN combinations. • Probabilistically assign whales to species based on features within the imagery and assessing manual ground truthing variably. • Development of a large-scale identification pipeline using virtual machines such as Google Earth Engine. • Assessing the use of UAV imagery as ground truthing and other possible verification techniques.
References - references should provide further reading about the project: 
Borowicz A, Le H, Humphries G, Nehls G, Höschle C, Kosarev V, et al. 2019. Aerial-trained deep learning networks for surveying cetaceans from satellite imagery. PLoS ONE. vol 14: e0212532. pmid:31574136
Höschle C. Cubaynes H.C. Clarke P.J. Humphries G. 2021. Borowicz A. The Potential of Satellite Imagery for Surveying Whales. Sensors. vol 21, pp 963.
Cubaynes HC, Fretwell PT, Bamford C, Gerrish L, Jackson JA. 2019. Whales from space: four mysticete species described using new VHR satellite imagery. Mar Mamm Sci. vol 35, 2 pp 466–491.
You can find out about applying for this project on the British Antarctic Survey (BAS) page.
Peter Fretwell
Dr Jen Jackson
British Antarctic Survey Graduate Administrator