An algorithm for aerosol optical properties retrieval over the ocean accelerated by a neural network from single-view multispectral measurements of intensity and polarization

Monitoring aerosols over the oceans is critical for understanding Earth’s climate and air quality. Although polarization can substantially reduce uncertainty in aerosol retrievals, current algorithms rely mainly on multiview polarimeters, and no dedicated algorithm is available for single-view polarimeters over the ocean. Here, we present the first ocean algorithm for a spaceborne single-view polarimeter, demonstrated with the particulate observing scanning polarimeter (POSP) onboard the GF-5(02) satellite. Our algorithm combines multispectral polarization with machine learning (ML)-accelerated radiative transfer (RT) calculation and seasonally clustered global aerosol models. Validation with Aerosol Robotic Network (AERONET) and maritime aerosol network (MAN) data demonstrates high accuracy, with RMSEs of 0.061, 0.479, and 0.037 for AOD550, aerosol loading (AE) 670−870 , and SSA550 using AERONET, and 0.030 and 0.259 for AOD550 and AE 670−870 using MAN, respectively. Comparison with retrievals from the generalized retrieval of atmosphere and surface properties (GRASPs) algorithm confirms that our algorithm performs comparably to GRASP products. These results underscore the necessity and feasibility of developing specialized aerosol retrieval algorithms for single-view polarimeters, and pave the way for global aerosol over the ocean monitoring.