Unsupervised Optical Classification of the Seabed Color in Shallow Oligotrophic Waters from Sentinel-2 Images: A Case Study in the Voh-Koné-Pouembout Lagoon (New Caledonia)

Guillaume Wattelez (1, 2) , Cecile Dupouy (3) , Farid Juillot (4)

1 UNC - Université de la Nouvelle-Calédonie
2 LIRE - Laboratoire Interdisciplinaire de recherches en éducation
3 MIO - Institut méditerranéen d'océanologie
4 IMPMC - Institut de minéralogie, de physique des matériaux et de cosmochimie

Monitoring chlorophyll-a concentration or turbidity is crucial for understanding and managing oligo- to mesotrophic coastal waters quality. However, mapping bio-optical components from space in such shallow settings remains challenging because of the strong interference of the complex bathymetry and various seabed colors. Correcting the total satellite reflectance signal from the seabed reflectance in ocean color with high resolution sensors is promising. This article shows how unsupervised clustering approaches can be applied to Sentinel-2 images to classify seabed colors in shallow waters of a tropical oligotrophic lagoon in New Caledonia. Data processing included Lyzenga correction for estimating the water column reflectance, optical spectra standardization for attenuating water absorption effects and clustering using the unsupervised k-means method. This methodological approach was applied on the 497, 560, 664 and 704 nm optical bands of the selected Sentinel-2 image. When applied on non-standardized data, our unsupervised classification retrieved three seafloor clusters, whereas five seafloor clusters could be retrieved using standardized data. For each of these two trials, the computed membership values explained more than 75% of the inertia in each Sentinel-2 wavelength band used for the clustering. However, the accuracy of the method was slightly improved when applied on standardized data. Confusion index mapping of the unsupervised clustering retrieved from these data emphasized the relevance and robustness of our methodological approach. Such an approach for seabed colors classification in optically complex shallow settings will be particularly helpful to improve remote sensing of biogeochemical indicators such as chlorophyll-a concentration and turbidity in fragile coastal environments.

Domaines: Machine Learning [stat.ML]; Traitement du signal et de l'image [eess.SP]

Guillaume Wattelez, Cecile Dupouy, Farid Juillot. Unsupervised Optical Classification of the Seabed Color in Shallow Oligotrophic Waters from Sentinel-2 Images: A Case Study in the Voh-Koné-Pouembout Lagoon (New Caledonia). Remote Sensing, 2022, 14 (4), pp.836. ⟨10.3390/rs14040836⟩. ⟨hal-03574120⟩