Retrieval algorithm for GEMS aerosol optical properties: Improvement, validation, and post-processing

Abstract

The Geostationary Environment Monitoring Spectrometer (GEMS), a pioneering Geostationary Earth Orbit (GEO) satellite instrument for air quality observation, is designed for atmospheric environmental monitoring and provides aerosol optical properties (AOPs). In this work, improvements to the GEMS aerosol retrieval algorithm, including spectral binning, surface reflectance estimation, cloud masking, and post-processing, are presented, along with validation results. These improvements aim to provide more accurate aerosol monitoring outcomes across Asia. Adopting spectral binning within the Look-Up Table (LUT) reduces random measurement errors and provides the stability of satellite data. Furthermore, Furthermore, a high-resolution surface reflectance database is constructed by considering monthly Background Aerosol Optical Depth (BAOD) values. This is based on the minimum reflectance method at the GEMS pixel resolution. The implementation of new cloud removal techniques enhances the accuary of cloud detection. Validation of GEMS AOD products against data from the AErosol RObotic NETwork (AERONET) and Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) from November 2021 to October 2022 reveals a robust correlation with AERONET AOD (R=0.792). Different validation outcomes are observed for different aerosol types, namely Highly Absorbing Fine, Dust, and Non-absorbing. GEMS Single Scattering Albedo (SSA) aligns well with AERONET data within acceptable error margins, although accuracy varies among aerosol types. When GEMS AOD exceeds 0.4, 42.76% of GEMS SSA values fall within the expected error range of ±0.03, and 67.25% fall within the range of ±0.05. The comparison of GEMS Aerosol Layer Height (ALH) with CALIOP data shows commendable agreement, with a mean discrepancy of -0.225 km and 55.29% (71.70%) of the data within ±1 km (±1.5 km). However, to address the issue of artifactual diurnal biases in AOD measurements, a machine learning-based post-processing correction method is developed. Post-process correction enhances GEMS AOD performance, reducing biases. In particular, the slope is close to 1, at 0.806 and the R is 0.897. Post-process correction also enhances GEMS SSA performance. 68.54% of GEMS SSA values fall within the expected error range of ±0.03, and 88.95% fall within the range of ±0.05.