Aerosol Effective Height Retrieval using O4 Absorption Property and Aerosol Classification Based on Machine Learning Approach from Space-Borne Hyperspectral Observation

Abstract

In this present study, an aerosol effective height (AEH) retrieval algorithm were developed using O4 absorption properties at 477 nm from the hyperspectral ultraviolet-visible space-borne sensors. In the AEH retrieval algorithm, an O4 slant column density (SCD) is retrieved using differential optical absorption spectroscopy (DOAS) technique, then O4 air mass factor (AMF) is derived by dividing O4 SCD by O4 vertical column density (VCD) to account for variation of O4 VCD associated with changes in atmospheric temperature and pressure. A spatiotemporal variation of O4 VCD and its effect on AEH retrieval accuracy was investigated. In addition, temperature-dependent cross section for O4 (TDCS) was applied to O4 AMF calculation. The AEHs were retrieved from satellite-based hyperspectral radiance from the ozone monitoring instrument (OMI) and tropospheric ozone monitoring instrument (TROPOMI) measurements. The retrieved AEHs were validated using both ground-based lidar measurement data. For the AEH retrieved from OMI measurement data (OMI AEH), a comparison between OMI AEH and those from ground-based lidar measurement data (lidar AEH) was carried out for the period from January 2005 to June 2009. A root mean square error (RMSE) between OMI AEH and lidar AEH found to be 0.44 km. The AEH from TROPOMI measurement data (TROPOMI AEH) was compared with lidar AEH for the period from January 2018 to June 2020. A difference between the TROPOMI AEH and lidar AEH is calculated to be 0.81 km. The difference between the validation result of the OMI AEH and that of the TROPOMI AEH may attributed to the low AOD condition (average aerosol optical depth (AOD) = 0.6) for the validation cases of the TROPOMI AEH. AEH retrieval errors were estimated using synthetic radiance simulated using a radiative transfer model. Uncertainties associated with input parameters (O4 VCD, O4 SCD, surface reflectance, TDCS, AOD, and aerosol type) for the AEH retrieval algorithm were considered in the error budget calculation procedure. Additionally, error budget was calculated under various AOD conditions. When AOD is increased, the AEH retrieval error tends to decrease, showing an increased sensitivity of the AEH retrieval algorithm. Uncertainties in O4 VCD and O4 SCD are found to have high contribution on the AEH retrieval accuracy. However, the contribution of uncertainties in TDCS and surface reflectance on the AEH retrieval are negligible. Misclassification of aerosol type found to have an effect on the AEH retrieval error up to 0.3 km, showing a contribution of accurate aerosol classification as the first step of the AEH retrieval algorithm while earlier satellite-based aerosol classification methods has not been evaluated. In this present study, we tried to evaluate the earlier aerosol classification methods and develop a new aerosol classification method based on machine learning approach. To classify aerosol type, satellite-based input variables are newly introduced. The new aerosol classification method is found to have sensitivity on identification of aerosol sphericity. Finally, the effect of the improved aerosol classification method on the AEH retrieval were investigated.