Using sentinel-2 satellite imagery to develop microphytobenthos-based water quality indices in estuaries

Remote Sensing
Ecology
Satellite
Sentinel
Machine Learning
First Author
Author

Simon Oiry; Laurent Barillé

Published

Feb 2021

DOI: 10.1016/j.ecolind.2020.107184

Microphytobenthos (MPB) is composed of unicellular photosynthetic organisms that colonize intertidal sediments within the first millimeters of the photic zone and form biofilms at low tide. In estuaries, this benthic group can represent the main primary producer and deliver several ecosystem services. However, it is not currently used as a bioindicator of water quality, contrary to the widespread use of phytobenthos in freshwater settings. This study assesses the potential of developing MBP metrics to assess water quality in transitional waters using Sentinel-2 (S2) satellite imagery. A Random Forest machine learning classification was used to distinguish different types of intertidal vegetation in 26 French estuaries and bays, in particular MPB and green macroalgae (chlorophytes), using multispectral indices. High accuracy was generally achieved for the identification of MPB when compared with field validation data, for both User’s and Producer’s accuracies, which corresponded to 94% and 84% respectively. Two Earth observation variables were retrieved: the Normalized Difference Vegetation Index (NDVI), a proxy of MPB biomass, and MPB percent cover integrated over the entire intertidal area of the water body. From a total of 918 S2 images from over a full year, 28% were exploitable due to the combined requirements of cloud-free pixels collected during low tide. With its 10 m spatial resolution, S2 was able to map all estuaries. MPB percent cover showed a stronger gradient between estuaries than MPB NDVI. MPB percent cover was also significantly correlated with green macroalgae percent cover, and a group of estuaries characterized by the highest MPB and green macroalgae coverage corresponded to eutrophic sites impacted by intensive farming activities. A multivariate analysis confirmed that MPB percent cover was indeed related to nutrients. It was also related to sediment type which was one of the main factors underlying differences between estuaries. This work is a first step toward a water quality metric using MPB, and several recommendations are proposed to refine this approach. Sentinel-2 imagery, which is publicly available, presents an interesting compromise to map estuarine microphytobenthos in order to assess the ecological status of transitional waters.