Validation of Sentinel-2 (MSI) and Sentinel-3 (OLCI) Water Quality Products in Turbid Estuaries Using Fixed Monitoring Stations

Abstract

It is common in estuarine waters to place fixed monitoring stations, with the advantages of easy maintenance and continuous measurements. These two features make fixed monitoring stations indispensable for understanding the optical complexity of estuarine waters and enable an improved quantification of uncertainties in satellite-derived water quality variables. However, comparing the point-scale measurements of stationary monitoring systems to time-snapshots of satellite pixels suffers from additional uncertainties related to temporal/spatial discrepancies. This research presents a method for validating satellite-derived water quality variables with the continuous measurements of a fixed monitoring station in the Ems Dollard estuary on the Dutch-German borders. The method has two steps; first, similar in-situ measurements are grouped. Second, satellite observations are upscaled to match these point measurements in time and spatial scales. The upscaling approach was based on harmonizing the probability distribution functions of satellite observations and in-situ measurements using the first and second moments. The fixed station provided a continuous record of data on suspended particulate matter (SPM) and chlorophyll-a (Chl-a) concentrations at 1 min intervals for 1 year (2016–2017). Satellite observations were provided by Sentinel-2 (MultiSpectral Instrument, S2-MSI) and Sentinel-3 (Ocean and Land Color Instrument, S3-OLCI) sensors for the same location and time of in-situ measurements. Compared to traditional validation procedures, the proposed method has improved the overall fit and produced valuable information on the ranges of goodness-of-fit measures (slope, intercept, correlation coefficient, and normalized root-mean-square deviation). The correlation coefficient between measured and derived SPM concentrations has improved from 0.16 to 0.52 for S2-MSI and 0.14 to 0.84 for S3-OLCI. For the Chl-a matchup, the improvement was from 0.26 to 0.82 and from 0.14 to 0.63 for S2-MSI and S3-OLCI, respectively. The uncertainty in the derived SPM and Chl-a concentrations was reduced by 30 and 23% for S2-SMI and by 28 and 16% for S3-OLCI. The high correlation and reduced uncertainty signify that the matchup pairs are observing the same fluctuations in the measured variable. These new goodness-of-fit measures correspond to the results of the performed sensitivity analysis, previous literature, and reflect the inherent accuracy of the applied derivation model.