Skip to main content

    Semi-automated software to count and validate Adélie penguin colonies from aerial photographs

    Request Meeting Document
    Document Number:
    WG-EMM-14/54
    Author(s):
    S.J. McNeill, K.J. Barton and P.O’B. Lyver (New Zealand)
    Submitted By:
    Mr Doug Cooper (CCAMLR Secretariat)
    Abstract

    Software has been written to enable semi-automated census counting and count validation of nesting Adélie penguins (Pygoscelis adeliae), on aerial photographs of the Ross Sea sector of Antarctica. The software is written in MATLAB®, is freely available and is deployed to users as a graphical user interface program. Previously, this task had been accomplished by manual marking of printed images, which is slow, and highly-dependent on the skill of the operator. The basis of the semi-automated counting procedure is linear discriminant analysis to separate the background (snow, water, rock, bare ground) from the guano-covered colony area, followed by morphological image processing operators to select the breeding penguins within the colony. Interactive features are provided in the software that allow an operator to add penguins omitted in the pattern recognition process, delete falsely-detected breeding penguins singly or in groups, selectively process a defined area, and record the running census counts. Validation of the counts against an experienced human counter is assisted by facilities to randomly sample the counted penguins, provide corrections to the census, and estimate the uncertainty of the counting process. The software has also been extended to allow for efficient counting of nesting Adélie penguins, although not automated pattern recognition, from historical black-and-white medium-format transparencies. We propose that our census software has much wider applicability than just counting penguins. While the present software has been written specifically for census counts of breeding Adélie penguins, there is no reason why the present approach could not be used for census counts of other species, such as seals on ice or farm animals on pasture. We outline the extent to which this software approach can be used with other species, and to aid the accuracy and reliability of count estimates from moderate-resolution satellite imagery.