Séminaire

This study presents a method for retrieving diagnostic pigments from satellite data. We trained a SOM map, which is an unsupervised neural network classifier, with in-situ HPLC diagnostic pigment concentrations collocated with GlobColour Chlorophyll-a concentration (Chl-a) and remote sensing reflectance, along with sea surface temperature measurements (SST) from AVHRR observed at global scale for a period ranging from 1997 up to 2014. After this training phase, the SOM map (denoted SOM-Pigment in the following) was able to retrieve the pigment concentrations in any ocean region using only satellite data (Globcolour data and SST) as input. These pigment concentrations were used to assess the presence of specific phytoplankton groups through pigment ratio determination. This method was applied to estimate the phytoplankton variability in the oligotrophic waters of the Mediterranean Sea.

The SOM-Pigment exhibited dissimilar performance depending on the type of pigment estimated in the Mediterranean. The coefficient of determination R2 ranged between 0.84 and 0.54 with an average RMSE of 0.016 mg.m-3. Each pigment showed coherent patterns and the distribution of phytoplankton size classes (PSC) was efficiently estimated. The Mediterranean showed a well-marked seasonal and geographical variability of Micro-phytoplankton (Micro 20-200µm), Nano-phytoplankton (Nano 2-20µm) and Pico-phytoplankton (Pico 0.2-2µm); the Nano dominated in the Winter-Spring months, and was more abundant in the western basin, while the Pico dominated in the Summer-Autumn months. The Micro was recurrent at the coast and in intense mixing zones, such as the Gulf of Lyon and the Rhodes Gyre, due to a sufficient nutrient input. Comparing these results with in-situ time series of phytoplankton counts proved that our approach efficiently reproduces the PSC spatio-temporal variability. Besides, a second methodology was built to identify dominant phytoplankton functional types (PFT) in the Mediterranean. For that, daily pigment images were generated for the Mediterranean via the SOM-Pigment and were used to build a new regionalized classification denoted SOM-PFT. As a result, six dominant PFT were identified (Diatoms, Synechoccocus, Prochlorococcus, Haptophytes, Chlorophytes, and Cryptophytes) and their spatio-temporal patterns was revealed. In-situ HPLC measurements sampled across the Mediterranean were used to validate the results of the SOM-PFT and recorded a general 20% error.

Consequently, these results show that the pigment reconstruction approach using the SOMs is suitable for basin and sub-basin scales, and allows monitoring the PSC and PFT variability.

julien.brajard@locean-ipsl.upmc.fr