Postdoctoral Research Associate in Plankton Imaging and Machine Learning (H/F)

- collect image data (UVP5, UVP6, Zooscan, FlowCam, ...) from international colleagues and assist in their upload to EcoTaxa (https://ecotaxa.obs-vlfr.fr )
- program and apply data quality control algorithms (e.g. outlier detection)
- develop and apply algorithms to evaluate the consistency of image classification across different image datasets
- evaluate and apply published size-to-biomass conversions for plankton and particles
- program data aggregation and transfer from EcoTaxa to PSSdb (API-based)
- combine image and environmental data (Temp., Oxygen, Satellite data) and use multivariate regression (e.g., generalized linear mixed models, generalized additive models) and machine-learning (e.g., RandomForest, neural networks) techniques to extrapolate data onto a global grid
- publish the results of these analyses in a scientific journal
- participate in international conferences and workshops to present the results of the project

Plankton is the major driver of oceanic biogeochemistry and plays several essential roles in the ocean (Oxygen production, Carbon storage, food web input, etc.). Plankton is the food base for fish and the livelihood of millions of people depends on plankton dynamics. Blooms of harmful algae or jellyfish on the other hand endanger aquaculture or even energy production. The rise of ocean optical technologies has yielded a diversity of underwater imaging tools (e.g. the Imaging Flow Cytobot, the Underwater Vision Profiler and the In situ Ichthyoplankton Imaging System) that enable the visualization of plankton and detritus ranging multiple orders of magnitude in size. Broad deployment of these imaging tools yielded datasets spanning the global oceans, yet few efforts have been made to bring these datasets together to evaluate and improve earth system models. The COMPLEx team at LOV, the Geophysical Fluid Dynamics Laboratory at Princeton University (USA) and the NOAA National Marine Fisheries Service (NMFS, USA: https://www.fisheries.noaa.gov/ ) are initiating a project to compile a global dataset of plankton and particle size structure, examine the biogeography of plankton/particle traits, and improve model-data intercomparisons for Earth System Models (ESMs).
We seek a postdoctoral researcher (0 – 1 year after the PhD) to work at the interface between marine ecosystem observations and machine learning. The ideal candidate would have a strong background in data management and machine learning approaches, with experience or interest in plankton imaging. We seek an individual proficient in scientific computing (e.g. Python, git, shell scripting) and with strong communication and project management skills. The individual will work with another postdoctoral researcher to be employed at GFDL and associated scientists, and will join an active group at LOV studying the biogeochemistry and ecology of the ocean. The researcher is expected to collate data from various locations and sources, focusing on data from five main imaging systems: Underwater Vision Profiler (UVP), Zooscan, In-situ Ichthyoplankton Imaging System (ISIIS), FlowCam, and Imaging Flow CytoBot (IFCB). He/She will develop a methodological framework for imaging data to be harmonized, compared against each other and collected in the Pelagic Size Structure database (PSSdb). The dataset should then be evaluated using different machine-learning approaches (niche-modelling, regression approaches, kriging, deep-learning) to assess the spatial and temporal representation and uncertainties associated with different regions and seasons and to establish patterns in the data that can be used for gap-filling.

The Laboratory of Oceanography of Villefranche-sur-mer (LOV; https://lov.imev-mer.fr/web/ ) is located close to Nice, on the French Riviera. It belongs to one of the three marine stations of Sorbonne Université. With about 90 permanent staff, the LOV generates and analyses a large quantity of marine data, including imaging, genomic, and satellite data to study the ocean.

The COMPLEx (COMPutational PLankton Ecology; https://lov.imev-mer.fr/web/team-complex/ ) team gathers ~30 members studying marine plankton and particles by collecting data with quantitative imaging instruments and high throughput genomics and using numerical data analysis methods (modeling, statistics, machine learning). COMPLEx strongly interacts with the Quantitative Imaging Platform of Villefranche (PIQv; https://sites.google.com/view/piqv ), which oversees the operation of tools that the team develops. Those tools include imaging sensors, such as the Underwater Vision Profiler or the ZooScan, and software packages, such as ZooProcess or the EcoTaxa web application (https://ecotaxa.obs-vlfr.fr/ ) that uses machine learning to assist taxonomists to sort plankton and particle images. We also have a long experience in interacting with computer scientists, in academia (e.g. LS2N in Nantes, I3S in Nice, ENSTA in Paris, MIP in Kiel), but also in the private sector (e.g. Google Brain lab in Paris).