Morphometrics and age from survey photographs
Ecological surveys are often photographed to allow for later analyses, which results in databases of images backlogged after serving their initial purpose. However, these images may contain additional valuable information, which can solve secondary objectives, even years after the original survey effort. I am currently developing a method to measure age and morphology from backlogged survey image databases, without the availability of a known subject distance or a scale in each image. African savanna elephants (Loxodonta africana) serve as an ideal model species to develop a non-invasive, image-based morphometric methodology: as handling these animals is particularly invasive and expensive, involving anesthetics. The Elephants Alive team, led by Dr. Michelle Henley, has conducted photographic surveys and has collared male elephants in the Associated Private Nature Reserves, South Africa (west of Kruger National Park) over the past 20 years. These images were originally used to identify individual elephants, based off of the elephants’ unique ear patterns, and are now being used to determine the tusk size, body size, and age of male elephants to better understand individual condition and whether hunting practices are sustainable.
Camera technology affords us the ability to deploy ‘virtual ecologists’ in hard-to-reach areas, or in places where human presence might disturb wildlife and therefore disrupt their behavior. By helping Dr. Tom Hart (PhD supervisor, University of Oxford) to expand a camera network in the Southern Ocean, a region of the world with harsh conditions year-round, I was able to observe novel behaviors and study penguin populations that have never before been surveyed because of their remote locations. Together, these approaches have resulted in an integrated monitoring network, which has the capacity to provide data to policy makers on areas particularly sensitive to fishing and human disturbance.
Along with the infinite possibilities of cameras as a monitoring tool comes an enormous amount of data in the form of hundreds of thousands of images. Each of the 72 cameras monitored by Dr. Tom Hart, his students (including myself until February 2017) and collaborators takes an image ranging from every hour to every minute daily, throughout the entire year. In order to turn this massive database of information into a data set, which can be used to answer hypotheses, the team has developed a citizen science webpage to crowd source data collection.
I worked with Zooniverse, an organization which provides scientists with an internet-based platform to connect with citizen scientists. The website that resulted, Penguin Watch, launched in September 2014 and asks users to annotate adults, chicks, eggs, and other animals present in each image. Users are presented with an image at random from a data set of roughly 500,000 images and 10 to 20 different users annotate each image. After multiple users have annotated an image, the Zooniverse team uses a clustering algorithm to combine the annotations so that each individual penguin receives one marking in the form of an x, y coordinate with a standard deviation.
The data is then compared to my own personal annotations as a gold standard to measure the accuracy of the citizen science annotations. Since the website’s launch, over one million images to date have been annotated by at least 2-10 users each, which has immensely sped up the data extraction process. Collaborators in Dr. Andrew Zisserman’s lab (Department of Engineering Science, University of Oxford) are also using volunteers’ annotations to create an algorithm, which automates the extraction of data from colonial seabird images using computer vision techniques.
Monitoring colonies over space and time
Using time-lapse images, I have studied the winter site fidelity of gentoo (Pygoscelis papua), chinstrap (Pygoscelis antarcticus), Adélie penguins (Pygoscelis adeliae) at multiple sites across the Southern Ocean. In addition, the cameras have provided information on the timing of distinct periods during the breeding period (eg. arrival and departure dates, hatching dates, incubation period, guard and post-guard periods) over multiple sites (up to 17) and years (up to 4), allowing for the study of patterns in phenology over space and time. By following individual nests throughout a breeding season, the cameras provide information on chick feeding rates and breeding success, which highlight vulnerable colonies and species.
Ultimately, I aim to answer hypotheses, which can directly feed into policy by determining important regions for penguins and highlight specific colonies of concern. There is growing support for Marine Protected Areas (MPA) within regions of the Antarctic, which would protect penguins by restricting fisheries activities. However, first, research must fill in gaps in our understanding of the species distributions and behaviors before implementing new conservation programs.