The main objectives of PHY-GO were to use concomitant measurements from cytometry, microscopy, and bio-optical sensors (e.g. Chl a fluorescence and optical backscatter, hereafter bbp) to develop an empirical model of phytoplankton community structure that could be applied to data collected by BGC-Argo floats—thereby providing insights into the partitioning of marine carbon at the global scale. The project consisted of four main objectives: 1) Constructing a matchup dataset of bio-optical, physical, and ecological data collected by the project sponsor and collaborators at Laboratoire d’Océanographie de Villefranche; 2) use machine learning to develop a model predicting phytoplankton community composition in response to bio-optical measurements alone; 3) evaluate relationships between phytoplankton community composition and variability in bio-optical measurements; 4) apply the model to better understand relationships between phytoplankton community structure and biogeochemical cycling in the global ocean.
All these objectives were largely completed during the fellowship period. Completion of the first objective produced a unique dataset consisting of more than 300 complete samples capturing phytoplankton dynamics across diverse locations, seasons, and depths. The PHY-GO dataset is one of the largest of its kind and provides a foundation for long-term advances in phytoplankton ecology and will support transformational research long beyond the end of the fellowship period.
Completion of the second objective yielded a novel neural network model providing predictions of heterotrophic bacteria, picophytoplankton (e.g. Prochlorococcus, Synechococcus, and picophytoeukaryotes) and nanophytoplankton. Although accurate predictions of microphytoplankton remain challenging due to their limited representation in the PHY-GO dataset, the model gives exceedingly high results in validation tests (e.g. R2 > 0.75 for all other groups), suggesting it is sufficiently robust to provide ecologically informative predictions. Interestingly, while relationships between gross community composition and bio-optical measurements were sufficiently strong to support predictions of community composition based on Chl a and bbp, phytoplankton diversity and taxonomy did not appear to contribute to variability in bio-optical measurements. Therefore, objective 3 was not completed beyond obtaining this null result.
Indeed, because the model relies only on pressure, temperature, salinity, Chl a, and bbp as inputs, it can be applied to an enormous amount of existing BGC-Argo data, thereby providing a strong foundation for the completion of objective 4. While work on this objective remains in its early stages, comparisons of 1) model predictions of phytoplankton community structure and 2) depth-resolved estimates of net primary production based on existing BGC-Argo data indicate an outsized role of nanophytoplankton in carbon cycling at the global scale, with small-sized diatoms emerging as key contributors to carbon cycling in high-latitude regions. Incorporation of physical measurements from BGC-Argo floats, biogeochemical models, and remote sensing platforms will provide further insight into the environmental drivers of these relationships.
