Aerosol Optical Depth Retrieval Based on Neural Network Model Using Polarized Scanning Atmospheric Corrector (PSAC) Data

Abstract

As the successors of the HuanjingJianzai-1 (HJ-1) series satellites in the Chinese Environmental Protection and Disaster Monitoring Satellite Constellation, the first two of HJ-2 A/B satellites have been successfully launched on the September 27 of 2020. The Polarized Scanning Atmospheric Corrector (PSAC) sensors, onboard the HJ-2 A/B satellites, are served as the synchronously atmospheric correction instrument requiring high speed and accurate aerosol optical depth (AOD) algorithm. For this purpose, we proposed a neural network based AOD retrieval model (named the AODNet), which takes full advantage of the multispectral measurements of PSAC for AOD retrieval with a high speed. The training of AODNet is conducted by the simulated observation data (currently applicable for the China region) from the forward calculation using the radiative transfer model. In this way, the land surface reflectance (LSR) is no need for our well trained model. It is expected to be one of the effective ways to solve the ill-pose problem in the decoupling of the atmosphere and surface information in AOD retrieval. Either of Sun-sky radiometer Observation NETwork (SONET) AOD or AErosol RObotic NETwork (AERONET) AOD was used to validate the AODNet AOD. The correlation coefficient is higher than 0.85 and more than 60% of the AODNet AOD can fall into the expected error envelope of ±(0.05+20%). The cross-comparison shows that the AODNet has better accuracy than MODIS Dark Target (DT) and Deep Blue (DB) algorithm. The air pollution episode is well characterized by the AODNet AOD using PSAC data.