I am most interested in using the newest advances in engineering (i.e. underwater camera systems for studying plankton) and data analysis (including deep learning and machine learning) to study the ocean and plankton ecology. In our current projects at the Hatfield Marine Science Center Plankton Ecology Lab we collect hundreds of millions to several billions of plankton images over a few weeks, using the ISIIS imager. While these billions of images have to be analyzed and automatically identified using vast computational resources which in itself is at the cutting edge of data science and oceanography, these new types of data (i.e., plankton distributions at centimeters to 1 m vertical resolution and over undulating transects of up to 100 km) give us the ability to tackle many ecological questions from new angles, and also open up completely new possibilities. One of my favorite things to do apart from just playing with those data, is visualizing it. For that I use a combination of ggplot2 in R, as well as Grafana.

For my PhD I worked on plankton ecology in the Canadian Arctic. Based on images from the LOKI underwater imaging system I developed an automatic identification model, utilizing machine learning. I then used automatically identified data to study diel vertical migration, seasonal vertical migration, individual lipid content of copepods, and how these features are coupled.

Prior to studying at Université Laval for my PhD I worked at the University of Alaska Fairbanks under Prof. Falk Huettmann where I started to develop my interest for machine learning algorithms and applying them to ecological questions. I started working on predicting mossy habitats on Haida Gwaii, BC, Canada which are invaluable for marbled murrelets (Brachyramphus marmoratus) as nesting sites. I then went on to write my actual M.Sc. thesis on predicting vertical and horizontal zooplankton distribution into the future. This was accomplished by analyzing a vast pan-Arctic dataset of physical oceanography parameters provided by Prof. Igor Polyakov and predicting oceanographic layers such as the mixed layer depth into the future. Using those layers as well as future data from the Canadian Earth System Model (CanESM2) RCP85 on chlorophyll concentration and similar I predicted various zooplankton features (e.g. preferred depth distribution, presence/absence) into the year 2100. For the modeling, the machine learning algorithms RandomForest and TreeNet were used.


I am generally interested in sea- and landscapes and how man-made disturbances affect them.


  • Optical imaging of planktonic organisms and particles in the oceans
  • Automated plankton image analysis using machine learning and deep learning
  • Spatial ecology and spatially explicit modeling
  • Zooplankton under changing climate, future climate predictions
  • Predator-prey interactions in the plankton
  • Society and nature: How is the status of the environment linked to economic well-being?

PICES 2019 presentation I recently gave in Victoria, BC, on our underwater imaging and deep learning efforts


Selected Research Projects

  • Effects of an upwelling mesoscale eddy on plankton distributions and predators prey interactions in the Straits of Florida

    This projects focuses on predator-prey interactions between the crustacean zooplankton and ichthyoplankton, and how they are affected by an upwelling eddy. Data for this project is derived from underwater imagery and deep learning

    See the presentation above for an introduction. As part of the OSTRICH project, an upwelling eddy in the Straits of Florida was located and adaptively sampled in June 2015. By carrying out undulating tows of up to 100 km length using the In-situ ichthyoplankton Imaging System (ISIIS), which at the same time resulted in plankton distributions at 1m vertical resolution, the propagating eddy was detected and multiple tows crossing from inside the eddy to the outside of the eddy were carried out. The so gathered vast quantities of images were automatically identified using a Sparse Convolutional Neural Network (this work was done by postdocs Jessica Luo and Kelly Robinson). Using these data we then identified tows on which clear eddy edges were present. On these transects interesting patterns were found. For instance, a close overlapping of Oithona sp. copepod hotspots with fish larvae hotspots was found inside the eddy, while outside of the eddy no such overlapping distributions were found (research ongoing). This research was funded by the  National Science Foundation (NSF).

  • Automated image analysis and a processing pipeline for very large amounts of underwater imaging data using Sparse Convolutional Neural Networks

    The Hatfield Plankton Ecology Lab is working on developing and further enhancing automated image analysis and a processing pipeline for the ISIIS underwater imaging system which collects 100s of millions to several billions of images over the course of a few weeks of research cruise

    This large and ongoing project has its roots in a Kaggle competition that the lab organized together with Booz Allen Hamilton in 2014/15 in order to find the best classifier for our complex problem of identifying many plankton groups and species from in-situ data.


    While we are currently working with the SparseConvNet solution by Benjamin Graham, described here: we are as well exploring the integration of the newer Submanifold Sparse Convolutional Networks, which can be found here:

  • Developing automatic species recognition for Arctic mesozooplankton using machine learning

    Based on detailed Lightframe Onsight Keyspecies Investigation (LOKI) plankton imagery (in-situ and without further subsampling), machine learning models were developed to automate species recognition

    This research was part of my PhD. Based on data collected during the ArcticNet 2013 cruise onboard the CCGS Amundsen, models for the automatic identification of copepod species and stages were developed. These automatic identifiers give us plankton distributions at a vertical resolution of 50 cm to 1 m across the water column and for about 30-35 different species/species groups. The model for 114 classes of plankton images was built using Random Forests:


    This research resulted in these papers:

  • Pelagic biodiversity components in the Southern Ocean under climate change

    Predicting spatial distributions of Antarctic pelagic fauna into the future

    Prof. Falk Huettmann and I worked on this project where we model-predicted spatial distribution of plankton, birds, and whales into the future using IPCC scenario data. The project highlights areas that will most likely experience the highest degree of change in the future regarding the occurrence of ecosystem critical species.

    From the chapter:

    “Our summary assessment of these findings suggests that species will show the greatest change in eastern Antarctic waters, Ross Sea region, Western Antarctic Peninsula, south of Australia, and around 60 degrees latitude.”


    The outcomes of this project were published in the  Biogeographic Atlas of the Southern Ocean, chapter 9.1. See at

  • Organizing research cruises

    While I participated in many oceanographic research cruises over the years, in the last three years I also played a major role in organizing research cruises with ~25 participating scientists each. This is a new and very interesting experience which made me realize that I like being in the role of organizing and facilitating research with many different partners/moving parts


    Cruise coordinator for the 2ndresearch cruise of the NSF-funded MEsoZooplankton trophodynamics in the CALifornia Current (MEZCAL) project onboard R/V Sally Ride. Sampling with a towed underwater imager and a coupled MOCNESS.



    Cruise coordinator for the 1stresearch cruise of the NSF-funded MEsoZooplankton trophodynamics in the CALifornia Current (MEZCAL) project onboard R/V Sikuliaq. Sampling with a towed underwater imager and a coupled MOCNESS.