POSTER SESSION 3
15:30–17:30, Wednesday, September 16
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
ABSTRACT 1196 | POSTER 139
SIMULTANEOUS RETRIEVAL OF DIAGNOSTIC PHYTOPLANKTON PIGMENTS FROM SENTINEL-3 OLCI USING CONVOLUTIONAL NEURAL NETWORKS
Phytoplankton pigment concentrations are key indicators of community composition, biogeochemical function, and ecosystem change. We present a multi-pigment ocean color retrieval framework for estimating 13 diagnostic phytoplankton pigments from Sentinel-3 OLCI-equivalent remote sensing reflectance measurements. Five machine learning architectures (Random Forest, XGBoost, dense neural network, one dimensional convolutional neural network, and bidirectional LSTM) were benchmarked using 185 coincident in situ reflectance and HPLC pigment measurements from optically diverse European waters. Pigments and reflectance were log transformed, split with output-aware stratification, and augmented using sample mixing and noise injection to represent radiometric and HPLC uncertainties. The convolutional neural network provided the strongest overall performance, with R2 values of 0.70 to 0.93 and mean absolute percentage errors of 25 to 55 percent across pigments. Performance for chlorophyll a, chlorophyll c1+c2, fucoxanthin, beta carotene, and diadinoxanthin approached reported interlaboratory HPLC uncertainty. To support satellite implementation, the OLCI-trained convolutional model was fine-tuned with a global matchup dataset combining open in situ pigment records with Copernicus GlobColour reflectances. Fine-tuning consistently outperformed training from scratch across all pigments, showing that high quality regional in situ radiometry can provide transferable spectral and biogeochemical features for global application contributing to an operationally viable path toward phytoplankton diversity products from multispectral ocean color observations.
Borja Sánchez-López, Institut de ciències del mar (ICM-CSIC), [email protected], https://orcid.org/0000-0002-8768-5422
Marco Talone, Institut de ciències del mar (ICM-CSIC), [email protected], https://orcid.org/0000-0002-9723-2080
Jesus Cerquides, Institut d’investigació en Intel·ligència Artificial (IIIA-CSIC), [email protected], https://orcid.org/0000-0002-3752-644X
Annalisa Di Cicco, Institute of Marine Sciences of the Italian National Research Council (ISMAR-CNR), [email protected]
Gonzalo Martínez-Fornos, Institut de ciències del mar (ICM-CSIC) and Universitat Politècnica de Catalunya (UPC), [email protected]
Petra Slavinec, National Institute of Biology, Marine Biology Station Piran, [email protected]
Poster Session | 1 | 2 | 3 | 4 | Schedule at a Glance
