A Machine Learning based approach towards products of Phytoplankton functional Types in the Arctic Ocean (ML-PhyTAO)

Both uncertainty assessment and validation have shown that the current global products of phytoplankton functional types (PFT) on Copernicus Marine Service for the Arctic Ocean (AO) bear larger gaps and higher uncertainties compared to that in the low latitude oceans. In the framework of Copernicus Marine Service Evolution Program, we propose a project ML-PhyTAO to exploit marine big data-driven machine learning (ML) methods in the PFT monitoring for high latitudes, and aim to set up an improved algorithm for better quantifications of multiple PFTs in the AO. A large marine data set (including bio-optical, biogeochemical, and physical data) obtained from various sources will be exploited as inputs for algorithm training and validation. The ML-PhyTAO is expected to deliver improved gap-free products of several key PFTs (diatoms, haptophytes, dinoflagellates, chlorophytes and prokaryotes) with uncertainty assessment to complement the current ocean colour/ biogeochemical data sets for the AO on the Copernicus Marine Service Data Store. Such PFT data set with improved accuracy will allow reliable long-term monitoring and trend analyses for the surface phytoplankton community structure, helping in detecting potential shifts and changes in phytoplankton diversity in the AO under the Arctic amplification effect. In this work we will demonstrate the framework of the project and present our latest outcome from the project by showing our first results on the experiments of ML methods using our well compiled training data sets including the in situ PFT data, satellite and model simulated data/products from Copernicus Marine Service covering various optical/physical/biogeochemical parameters.