Sensitivity Assessment of Atmospheric Corrections for Clear and Moderately Turbid Optical Waters

Atmospheric Correction (AC) aims to restore surface reflectance from the top of the atmosphere (TOA) reflectance. In this study, seven highly established state-of-the-art AC (OC-SMART, ACOLITE, Polymer, Sen2Cor, iCOR, C2RCC, and L2gen from SeaDAS) approaches were employed on Sentinel-2 MSI high-resolution imageries. The performance of the AC algorithms was validated by comparing the satellite-derived remote sensing reflectance ($R_{rs}(\lambda)$) at four visible wavelengths (443, 460, 590, and 665 nm) with the co-located in-situ hyperspectral measurements acquired within a temporal window of $\pm $3 hours across various water zones located in the Atlantic Ocean. 75 optimal match-up pairs were obtained from six sites between December 2018 and August 2020. An unsupervised learning technique was used to classify the in-situ hyperspectral $R_{rs}(\lambda)$ measurements into three optical water types encompassing nearly clear to moderately turbid coastal and deep water zones. Upon general analysis, OC-SMART produced the most precise $R_{rs}(\lambda)$ with and $S_{tot}$ score and $\bar{\chi }^{2}$ value of 20.54/24, and 0.11, respectively. L2gen AC produced $R_{rs}(\lambda)$ that show the highest resemblance with the spectral shape in terms of Spectral Angle and Quality Assurance score with the in-situ $R_{rs}(\lambda)$ that are 10.4 and 0.84, respectively. The performance of ACs varies across the water types and wavelengths. Using existing bio-optical algorithms, the validation is further extended by obtaining downstream water-quality parameters, such as $Chl_{a}$, TSS, and $a_{cdom}(440)$, from the in-situ measured and atmospherically corrected $R_{rs}(\lambda)$. The expected reasons that affect the performance of ACs across designated water types were discussed.