Phytoplankton play a key role in the regulation of numerous biogeochemical cycles. Their role varies depending on their biomass, but also on their community structure. Therefore, monitoring phytoplankton biomass and community structure over large spatial, vertical and temporal scales is a key challenge. BGC Argo floats, equipped with fluorometer, hydrological and optical sensors, provide a great opportunity to develop new methods for assessing phytoplankton biomass and community structure. In parallel, the recent development of machine learning methods makes it increasingly possible to derive biological and ecological variables from a set of environmental variables. We have therefore assembled a dataset of simultaneous measurements of phytoplankton biomass and community composition with BGC-Argo variables. In this study, we present how we can use two different machine learning methods to improve phytoplankton biomass and community structure from BGC-Argo float observations. The first one allows to infer particulate organic carbon from the standard BGC-Argo variables, making it potentially applicable to the whole BGC-Argo fleet. The second uses in addition the particle beam attenuation measured on a large number of floats, which allows to discriminate the particulate organic carbon stock into four different plankton groups, i.e. bacteria, pico-, nano- and micro-phytoplankton.  Finally, the combination of these two methods allows the organic carbon stock of these four plankton groups to be estimated from BGC Argo floats.

Primary Presenter: Flavien Petit, Sorbonne Université (flavien.petit@imev-mer.fr)

Authors:

Julia Uitz, CNRS - Sorbonne Université  (julia.uitz@imev-mer.fr)

Hervé Claustre, CNRS - Sorbonne Université  (herve.claustre@imev-mer.fr)