Why aerial imagery and AI are important for cetacean monitoring and conservation

Aerial view of narwhals (highlighted in green).

Whales live in all our oceans and seas, from the high arctic to Antarctica and spend most of their time underwater. Monitoring whale populations on such large scales is a challenging task for scientists, conservationists, and managers. One proven monitoring method is aerial surveys. Compared to surveying marine mammals from boats, surveyors aboard aircraft can cover much larger areas and reach remote and often inaccessible locations far offshore or covered in scattered ice floes.

Bertrand at work during an aerial survey.

Despite these benefits, aerial surveys require a lot of financial and human resources. Multiple marine mammal observers are needed to detect, identify, and count individual whales. From firsthand experience onboard an aircraft, this is not an easy task when you’re flying at 100 knots/hour at 1000 or 2000 feet! As an aerial observer, you would often be happy just to have identified the species observed and to have had a somewhat accurate number of whales present at the surface.

This is where aerial imagery becomes extremely valuable. Still images not only allow us to count surfacing whales more accurately—images also give us the opportunity to retrieve more information about group composition, such as the number of adults, juveniles, and newborns. Using these detailed data, scientists can better understand and predict future population decline or growth, based on the number of reproductive females present for instance. They can also identify key habitats for mating or calving, which can help managers prioritize hotspots for conservation policy. Moreover, imagery allows multiple observers to review an uncertain observation to decide on the identification of a target.

So why don’t people use aerial imagery more often? While aerial images can be a valuable source of information the number of images collected over a survey is the main pain point to leverage this data source. We are talking of hundreds of thousands of images to process. Analyzing these images by hand can take a few minutes to a couple of hours per image depending on the complexity of the images (i.e., glare, sea state, ice) and the number of targets found. Manual annotation is currently the status quo for aerial surveys, preventing aerial monitoring from being a truly scalable solution.

To solve the scalability issue of aerial imagery, we urgently need AI solutions to process these data in a short period of time. I want to make a distinction between AI and an AI solution because many people mistakenly think that you need only to develop an AI algorithm and then you are set for life. An AI solution is not a hands-free solution. Any AI algorithm that you develop needs to be integrated into a pipeline infrastructure that can be fed large amount of data via different sources and formats, and output different formats if necessary. If you don’t have a flexible infrastructure, then your algorithm will have very limited use. Moreover, you need to ensure that your model doesn’t lose its ability to recognise a feature overtime (a phenomenon known as model drift). Thus, you need to constantly guide and retrain your model to keep it running at a high calibre.

Manual whale detection annotations (yellow) versus Whale Seeker’s whale detection algorithm (red)

At Whale Seeker, we’ve put a lot of time and effort into developing the support infrastructure of our AI model, which results in a very powerful, scalable, and standardized solution capable of rapidly and consistently processing terabytes of data. This means that we can now more easily compare results over time and track changes in a population.

Conducting aerial surveys and collecting data is a costly endeavour that is not currently utilized to its full potential. We should put a similar amount of focus and investment into the analysis of aerial data to gain as much information as possible from these efforts.  AI solutions have an upfront cost but the return on investment rises quickly over a short period of time. With the rapidly increasing scale of maritime industries and ocean management projects, AI solutions are becoming an essential component for the monitoring marine mammal populations.