Phytoplankton Diversity in the Mediterranean Sea From Satellite Data Using Self-Organizing Maps

We present a new method to identify phytoplankton functional types (PFTs) in the Mediterranean Sea from ocean color data (GlobColour data in the present study) and AVHRR sea surface temperature. The principle of the method is constituted by two very fine clustering algorithms, one mapping the relationship between the satellite data and the pigments and the other between the pigments and the PFTs. The clustering algorithms are constituted of two efficient self-organizing maps, which are neural network classifiers. We were able to identify and estimate the percentage of six PFTs: haptophytes, chlorophytes, cryptophytes, Synechococcus, Prochlorococcus, and diatoms. We found that these PFTs present a peculiar variability due to the complex physical and biogeochemical characteristics of the Mediterranean Sea: Haptophytes and chlorophytes dominate during winter and mainly in the western Mediterranean basin, while Synechococcus and Prochlorococcus dominate during summer. The dominance of diatoms was mainly observed in spring in the Balearic Sea in response to deep water convection phenomena and near the coastline and estuaries due to important continental inputs. Cryptophytes present a weak concentration in the Aegean Sea in autumn. The validation tests performed on in situ matchups showed satisfying results and proved the ability of the method to reconstruct efficiently the spatiotemporal patterns of phytoplankton groups in the Mediterranean Sea. The method can easily be applied to other oceanic regions. Plain Language Summary The identification and spatiotemporal distribution of phytoplankton assemblages give powerful insights on the dynamics of the marine food web and the ocean role in climate regulation in the context of the global change. A new method to identify phytoplankton functional types from satellite observations has been developed and applied in the Mediterranean Sea. It is based on artificial intelligence clustering, the so-called self-organizing maps. The method was able to differentiate multiple phytoplankton assemblages and to provide their different pigment compositions. This approach had been validated successfully with in situ data sets and the spatiotemporal variability of the phytoplankton functional types showed a significant coherence. The method is very general and can be applied to other regions.