Application of Random Forest-based Aerosol Classification Model on Aerosol Height Retrieval from Space-borne Hyperspectral Measurements

The accuracy of the O4-based aerosol height retrieval algorithm can be affected by uncertainties of input parameters (aerosol type, aerosol optical depth, and surface reflectance, etc.). In this study, we attempted to improve the O4-based aerosol height retrieval algorithm by applying a new aerosol type classification method. At first, we investigated the effect of aerosol misclassification on aerosol height retrieval using synthetic radiance from the simulation of the radiative transfer model (Linearized pseudo-spherical scalar and vector discrete ordinate radiative transfer models; VLIDORT). Aerosol height retrieval error due to aerosol misclassification was calculated to be ~ 0.3 km. Secondly, the satellite aerosol classification method was newly developed based on the Random Forest model and applied to the aerosol height retrieval algorithm. By applying with the new aerosol classification model, the aerosol height was retrieved from the low orbit environmental satellite observations (Ozone Monitoring Instrument; OMI, TROPOspheric Monitoring Instrument; TROPOMI). The new application results were evaluated using the lidar measurement data over Asia (Asian Dust and Aerosol Lidar Observation Network). Additionally, the effect of the new satellite classification model on aerosol height retrieval was investigated by comparing it with the use of a previously developed satellite classification model.