In my PhD project, we work with recently developed underwater microscopes that take in situ photos of zooplankton. The number of images we collect in this way far exceeds what is possible for a human to sort manually, therefore our focus is to develop and apply machine learning techniques that can identify and count plankton species from such images automatically. For this we make use of Convolutional Neural Networks, a form of deep-learning that has become very popular over the past decade for all kinds of image recognition applications.
Plankton comes from a Greek word meaning ‘to wander, to drift’ and refers to all pelagic organisms in the sea that are unable to swim against currents. Zooplankton are the animal-like groups therein and they play a vital role in the marine pelagic food web, being the link between primary production from algae up to higher organism such as fish, sea birds and mammals. Using these plankton imaging instruments, we can obtain fine-grained data on zooplankton abundance in time and space, and linking these data to seawater conditions such as temperature, depth, salinity and amount of dissolved oxygen, enables us to discover patterns that would be impossible to find with traditional zooplankton sampling methods using nets, since these are much more limited in their temporal and spatial resolution.
Zooplankton is particularly sensitive to changing environmental conditions - there are species distributions that have shifted northwards with 200 kms per 10 years over the past decades - and given the important role in marine pelagic ecosystems, some past shifts in marine ecosystems have been linked to actual changes in zooplankton communities. Examples of these are sudden collapses of fish stocks or a changed regional occurence of whales. And with more human-induced changes expected to occur in the Dutch North Sea in the coming decades, such as warming of sea water and construction of offshore wind farms, it is therefore especially important to closely monitor zooplankton communities in this region, and to understand how these communities are linked to, for example, primary production and environmental conditions. Currently though, this is not being monitored on a structural basis.
One of the goals of this PhD project is to facilitate future monitoring of zooplankton in the North Sea using plankton imaging instruments. In addition to programming and investigating new deep-learning techniques, I also spend my time by participating in various research cruises within the Zeeland region and the North Sea in order to collect new data and to do novel ecological research using these exciting techniques.
In the past, development of observational methods have often gone hand in hand with new scientific insights, so I am curious to what the coming years will bring us.