Retrieving oceanic constituents and inherent optical properties in global oceans using multi-angular polarimetric...

Multi-angle polarization characteristics of water-leaving radiation are crucial for the retrieval of the oceanic constituents and the inherent optical properties (IOPs). However, the feasibility of using the polarization spectrum to retrieve the oceanic constituents and IOPs in global oceans remains unclear. In this study, global radiative transfer (RT) simulations for the Stokes vectors of water-leaving radiance (L_w) were performed, and a global polarization-based algorithm using the fully connected U-Net (FCUN) for retrieving the oceanic components and IOPs was developed. The performance of the FCUN algorithm was compared with in situ measurements collected at Qiandao Lake, China. The results indicated that the degree of polarization (DOP) at 443 nm in the global ocean waters was low in oligotrophic waters, with a minimum of 0.0176. The mean absolute percentage errors (MAPEs) of the FCUN algorithm for retrieving various oceanic constituents and IOPs at 443 nm were low with a range of 3.56-16.97%. Moreover, the predictions of the FCUN algorithm were consistent with the RT simulation inputs under conditions with various random instrument noise; the MAPE values were 6.74% and 8.84% for the oceanic constituents and IOPs, respectively. Furthermore, the performance of the FCUN algorithm on the field measurements was validated; the MAPE values for chlorophyll-a, the total absorption coefficient at 443 nm, the absorption coefficient of phytoplankton at 443 nm, and the total backscattering coefficient at 443 nm were 31.80%, 29.65%, 34.87%, and 43.04%, respectively. Additionally, the importance of multi-angle polarized L_w observations for retrieving the global ocean constituents and IOPs was also examined, and the MAPE values decreased from 16.91% to 1.48% as the observation angles increased. Overall, the polarization-based inversion model has substantial potential for the retrieval of oceanic constituents and IOPs.