Improving Ocean Color Atmospheric Correction Near Clouds With Neural Network

Atmospheric correction is a prerequisite for deriving accurate ocean color products from satellite observations. However, current atmospheric correction algorithms directly mask clouds and around area to avoid the influence of cloud adjacency effects (CAEs), which degrade the quality of satellite observations, limiting the spatial-temporal continuity of water quality monitoring. In this study, we introduce the CAEs-effected Atmospheric correction for Polar-orbit satellites (CAP) algorithm, demonstrated using MODerate resolution Imaging Spectroradiometer-Aqua (MODISA) as an example. CAP is a physically constrained data-driven approach designed to enhance atmospheric correction performance in the presence of CAEs. The algorithm integrates the physical foundation of the standard atmospheric correction scheme implemented in SeaDAS with the cloud-adjacency correction information extracted from GOCI data. The GOCI-based CAE features are transferred to MODISA spectral bands and jointly used for model training. Validation using independent testing datasets shows that CAP achieves band-averaged correlation coefficients (R) of 0.97 with minimal bias and significantly reduces relative percentage differences near cloud edges. CAP improves signal-to-noise ratios by up to 53.7%, particularly in the red bands, and increases the number of valid observations by ~40%. Global-scale composites confirm that CAP can generate Rrs fields consistent with the results of the standard atmospheric correction in SeaDAS without applying straylight masks, thereby recovering valuable observations previously discarded. By mitigating CAE-induced biases while maintaining accuracy in cloud-free conditions, CAP enhances the reliability of bio-optical retrievals and expands observational coverage for ocean color applications. The methodology of integrating geostationary and polar-orbit sensor data provides a transferable framework for future satellite missions and can be extended to other challenging atmospheric correction scenarios.