Using high resolution satellite data and AI to quantify changes in penguin populations

Past studies have shown it is possible to detect and count birds using drone, satellite or time lapse video camera traps [1,2,3], often utilising citizen science or machine learning approaches to processing images. This project will aim to further such work, using machine learning approaches to satellite or drone data to perform instance segmentation in order to estimate penguin populations in Antarctica. Also there is an aim to expand this methodology to be able to work at scale and be adaptable to work on a variety of species. Alongside this the project will seek to develop a robust framework for incorporating that analysis, and the associated error rate and structure into a broader statistical population model. Additionally it is of interest to look at combining several forms of data (satellite, drone, nest-based video capture) into one integrated analysis, with the aim of better understanding the errors associated with each, and better understanding how to constrain and improve the more cost effective of these strategies which are often intrinsically less accurate.