Remote Sensing of Coastal Water-quality Parameters from Sentinel-2 Satellite Data in the Tyrrhenian and Adriatic Seas

The Sentinel-2 mission allows the remote sensing of coastal waters environments with detailed imagery down to 10-m resolutions. In this study we explore different approaches in the retrieval of chlorophyll-a (Chl-a) and total suspended matter (TSM) concentrations along the Adriatic and Tyrrhenian coasts in Italy, using empirical and model-based data to develop regressive and neural networks (NN) algorithms. The best performances, for both Chl-a and TSM, are obtained with NN algorithms, better suited in handling the high variance shown by in-situ target data, with a 0.33 mg m
⁻³
root-mean-square-error (RMSE) for Chl-a and
$\mathrm{RMSE} =1.95\mathrm{g}/\mathrm{m}^{3}$
for TSM. The forward bio-optical model used for the generation of synthetic training data shows inadequacies in modelling the different spectral features characterizing the oligotrophic Tyrrhenian Sea and the eutrophic Adriatic Sea. The suitability of Sentinel-2 MSI products is here confirmed, but further investigation is needed on some critical issues affecting remote sensing of coastal waters, such as the atmospheric correction problem, the analysis of bathymetric effects and the need of a large and reliable in-situ dataset.