Due to the complexity of water and atmosphere properties, it has been a long-standing challenge to accurately remove atmospheric effects for waters in coastal regions. In this study, using a large collection of remote sensing reflectance (Rrs) in coastal waters obtained by the Aerosol Robotic Network-Ocean Color (AERONET-OC), we developed a neural-networks based algorithm specifically for the atmospheric correction of coastal waters (NNAC_CW). For more than 10,000 matchups between MODIS-Aqua and AERONET-OC measurements covering a wide range of water types, it is found that the Rrs of MODIS-Aqua obtained by NNAC_CW are in very good agreement with the Rrs from AERONET-OC, where the coefficient of determination (R²) is in a range of ~0.73-0.90 for wavelengths of 412 – 678 nm, with the mean absolute percentage difference (MAPD) in a range of 15% – 39%. In contrast, the values of R² and MAPD are in a range of 0.51 – 0.89 and 23%-74%, respectively, between the Rrs of MODIS-Aqua and the Rrs from AERONET-OC when data of MODIS-Aqua were processed following the conventional scheme for atmospheric correction. These results indicate a promising potential to use Rrs data from AERONET-OC as a base to develop robust atmospheric correction algorithm for coastal and inland waters.

nerual network;,ocean color remote sensing,coastal zone