The dataset of water-leaving radiance polarization components includes the atmospheric diffuse transmittance lookup table (PLUT), aerosol models, aerosol optical thickness, and Stokes components above the water surface (Iw, Qw, Uw) corresponding to open ocean areas globally (60°S–60°N). The units for Stokes components are per steradian (sr-1). The spatiotemporal resolution and coverage of this dataset depend on the input dataset. For example, the spatial resolution of the PARASOL polarized satellite data is 6 km2, covering global ocean areas every two days from March 2005 to October 2013. Verification of the off-water radiance polarization component dataset using the accurate vector radiative transfer model OSOAA indicates a root mean square error of approximately 10-4 sr-1.
The dataset mainly includes the calculation results of two inversion models, namely PACNIR and IPAC models. The PACNIR model uses polarized remote sensing data (It, Qt, Ut) at the top of the atmosphere over the target ocean region as algorithm input. It employs a nonlinear optimization algorithm (Nelder-Mead simplex algorithm) to construct a system of equations for all observed directions and solves them simultaneously. Based on a pre-built lookup table of atmospheric diffuse transmittance for water-leaving radiance polarization components, it generates aerosol models, aerosol optical thickness, and polarized radiance component values above the water surface for the target ocean region. Compared to traditional atmospheric correction products that are designed for single-angle, scalar ocean color satellite data (such as MODIS, VIIRS), this product represents the first attempt to perform atmospheric correction calculations specifically for polarized satellites, resulting in polarized water-leaving radiance component values for open ocean areas. This product can be used for subsequent studies on ocean-atmosphere interactions and target detection based on polarized signals. The IPAC model further improves the slow computation speed of the PACNIR model by using a machine learning algorithm (XGBoost). The IPAC model utilizes global distributions of satellite data products such as chlorophyll concentration, coarse- and fine-mode aerosol optical thickness, coarse- and fine-mode aerosol complex refractive index, sea surface wind speed, yellow substance absorption coefficient, and suspended particulate matter concentration. Based on vector radiative transfer simulations, a large number of Stokes values of the top of the atmosphere and the bottom of the atmosphere are calculated. The correlation between these signals is established and the computational time required for atmospheric correction of the water-leaving radiance is significantly reduced. These efforts lay a solid foundation for the further operational production and application of polarized water-leaving radiance products.
