Enhancing the usability of Satellite Earth Observations through Data Driven Models. An application to Sea Water Quality

Earth Observation from satellites has the potential to provide comprehensive, rapid and inexpensive information about land and water bodies. Marine monitoring could gain in effectiveness if integrated with approaches that are able to collect data from wide geographic areas, such as satellite observation. Integrated with in situ measurements, satellite observations enable to extend the punctual information of sampling campaigns to a synoptic view, increase the spatial and temporal coverage, and thus increase the representativeness of the natural diversity of the monitored water bodies, their inter-annual variability and water quality trends, providing information to support EU Member States’ action plans. Turbidity is one of the optically active water quality parameters that can be derived from satellite data, and is one of the environmental indicator considered by EU directives monitoring programmes. Turbidity is a visual property of water, related to the amount of light scattered by particles in water, and it can act as simple and convenient indirect measure of the concentration of suspended solids and other particulate material. A review of the state-of-the-art shows that most traditional methods to estimate turbidity from optical satellite images are based on semi-empirical models relying on few spectral bands. The choice of the most suitable bands to be used is often site and season specific, as it is related to the type and concentration of suspended particles. When investigating wide areas or long time series that include different optical water types, the application of machine learning algorithms seems to be promising due to their flexibility, responding to the need of a model that can adapt to varying water conditions with smooth transition, and their ability to exploit the wealth of spectral information. Moreover, machine learning models have shown to be less affected by atmospheric and other background factors. Atmospheric correction for water leaving reflectance, in fact, still remains one of the major challenges in aquatic remote sensing. The use of machine learning for remotely sensed water quality estimation has spread in recent years thanks to the advances in algorithm development, computing power, and availability of higher spatial resolution data. Among all existing algorithms, the choice of the complexity of the model derives from the nature and number of available data. The present study explores the use of Sentinel-2 MultiSpectral Instrument (MSI) Level-1C Top of Atmosphere spectral radiance to derive water turbidity, through application of a Polynomial Kernel Regularized Least Squares regression. This algorithms is characterized by a simple model structure, good generalization, global optimal solution, especially suitable for non-linear and high dimension problems. The study area is located in the North Tyrrhenian Sea (Italy), covering a coastline of about 100 km, characterized by a varied shoreline, embracing environments worthy of protection and valuable biodiversity, but also relevant ports, and three main river flow and sediment discharge. The coastal environment in this area has been monitored since 2001, according to the 2000/60/EC Water Framework Directive, and in 2008 EU Marine Strategy Framework Directive 2008/56/EC further strengthened the investigation in the area. A dataset of combination of turbidity measurements, expressed in nephelometric turbidity units (NTU), and values of the 13 spectral bands in the pixel corresponding to the sample location was used to calibrate and validate the model. The developed turbidity model shows good agreement of the estimated satellite-derived surface turbidity with the measured one, confirming that the use of ML techniques allows to reach a good accuracy in turbidity estimation from satellite Top of Atmosphere reflectance. Comparison between turbidity estimates obtained from the model with turbidity data from Copernicus CMEMS dataset named ’Mediterranean Sea, Bio-Geo-Chemical, L3, daily observation’, which was used as benchmark, produced consistent results. A band importance analysis revealed the contribution of the different spectral bands and the main role of the red-edge range. Finally, turbidity maps from satellite imagery were produced for the study area, showing the ability of the model to catch extreme events and, overall, how it represents an important tool to improve our understanding of the complex factors that influence water quality in our oceans.