Exploring Machine Learning for use in Seabird Monitoring

Research leader: Aili Labansen, Researcher, Greenland Institute of Natural Resources

Other participants: Flemming Merkel, Senior Researcher, Greenland Institute of Natural Resources and Aarhus University; Paula Schmidt, MSc in Biology

Research area, purpose, and research question:
At  Greenland Institute of Natural Sciences we incorporate photography in connection with our monitoring program of the breeding population of Arctic terns (appa) and terns (taateraat) in Greenland. The monitoring program gives us, among other things, knowledge of the status of the breeding stocks, changes over the breeding season and breeding success.
We work with two types of images: photography of entire colonies, for the purpose of total counts, and time-lapse images of a sample of specific colonies, taken by fixed cameras at regular intervals throughout the breeding season. The time-lapse images provide data for when the breeding season starts and ends, the variation in the number of birds in the colony over the whole season and the proportion of breeding birds that have young on their wings.
Fieldwork in bird colonies is demanding and expensive. The use of photography has greatly improved surveillance with increased accuracy and by enabling the collection of data that would otherwise be too time-consuming and costly. The downside is a significant increase in workload back in the office. A problem that machine learning could potentially help solve.
With this project, we are seeking funding to employ a recently graduated biologist with experience in machine learning and image analysis. Our goal is to investigate whether it is possible to use existing image material for automated identification and counting of birds based on digital images from our monitoring program. The new methods will be able to be verified with results obtained with our current methods.

Methods for capacity building in Greenland:
By hiring Paula to help us develop methods to make our monitoring program more efficient, we leverage our own skills in machine learning and image analysis, which can potentially improve our monitoring program.

Dissemination plan:
A podcast about the method development and about our monitoring program in general will be prepared in collaboration with our communications department, as popular scientific communication.

Granted: DKK 100.000