Aerosol Fine-Mode-Fraction Retrieval from GEO-KOMPSAT-2A/AMI using a Deep Neural Network and Spectral Deconvolution Algorithm

In the Geostationary Korea Multi-Purpose Satellites (GEO-KOMPSAT; GK) mission, South Korea has launched two geostationary earth orbit satellites, GK-2A and GK-2B. GK-2A is equipped with the Advanced Meteorological Imager (AMI), which spans spectral bands from visible to infrared. GK-2B carries two sensors, one of which is the Geostationary Environment Monitoring Spectrometer (GEMS), an ultraviolet-visible hyperspectral spectrometer. Each sensor's distinct characteristics, derived from their individual specifications, are crucial for retrieving aerosol optical properties. Aerosol size information, a critical factor in understanding aerosol effects on human health, can be quantified by the Ångström Exponent (AE), which describes the spectral dependence of aerosol optical depth (AOD). AE's inverse correlation to average aerosol size makes it a valuable metric for assessing aerosol size information from satellite data. Instruments with more than one VIS–NIR channel can easily calculate AE from spectral AOD. In our study, we employed data fusion of level 2 aerosol products and level 1B radiance observations from GEMS and AMI using a deep neural network (DNN). We trained a DNN model on the ground-based AErosol RObotic NETwork (AERONET) dataset, covering the period from November 2021 to October 2022, for initial retrieval of AOD and AE. Subsequently, we converted spectral AOD to fine mode fraction (FMF) using a spectral deconvolution algorithm. Results show that the DNN-derived spectral AOD, AE, and FMF closely match the AERONET dataset, during the period not used in the DNN model training. Additionally, a case study on Asian dust transport demonstrated the utility of synergistic fusion of multiple geostationary Earth orbit satellite instruments. This study highlights the potential of the DNN data fusion to enhance aerosol property retrieval, providing valuable insights into aerosol dynamics in East Asia.