Atmospheric Correction Based on Principal Component Analysis in Highly Turbid Waters of the Río de la Plata Estuary

Monitoring of water quality by satellite ocean color data requires accurate atmospheric correction. The treatment of the atmospheric signal becomes more complicated in turbid coastal regions, where the usual black pixel assumption in the Near Infra-Red (NIR) correction bands is oftenly invalid due to high backscattering from suspended substances present in the water. One of the usually proposed solutions is to assign correction bands further apart in the spectrum: the Short-Wave Infra-Red (SWIR) bands, where the black pixel assumption still holds in turbid waters due to higher water absorption. But in exceptionally turbid regions, such as in the Río de la Plata estuary, where suspended particulate matter concentration can surpass 400 mg/L, the marine signal at 1240 nm may exceed 0.2%, that is, not black for the purpose of atmospheric correction. Moreover, as the spectral distances between the correction bands and the NIR region increases, the correlation between the atmospheric signals of the former with the latter ones usually worsens, threatening the accuracy of the atmospheric correction.

In this work, an atmospheric correction scheme based on Principal Component Analysis of the TOA signal and a simple expression for the transmission coefficient was proposed to estimate water reflectance in the NIR region using four different sets of SWIR correction bands present in MODIS and in the Argentinian-Brazilian future sensor SABIA-Mar. Two factors changed between the four sets: the amount of bands used (2 or 3 bands) and the spectral distance to the NIR, which regulates simultaneously the degree of correlation and the validity of the black pixel assumption.

The atmospheric correction performances were evaluated from a set of simulations of TOA signals using the CNES-SOS radiative transfer code and in situ data available for the Río de la Plata region. All four schemes present better performances when low observation and solar zenith angles (i.e. at low air masses) and low aerosol optical depths occur, being the 1640-2130 band combination the one that performs better globally. Finally, the atmospheric correction scheme was applied to a MODIS image in the Río de la Plata estuary and the water reflectances obtained were compared with in situ radiometric measurements taken simultaneously with the image.