The optimal single-scattering properties for retrieving ice cloud properties based on machine learning techniques

Ice clouds play significant roles in the climate system by modulating radiation budget and hydrology cycle. It is essential to robustly retrieve ice cloud properties by using satellite measurements. In the existing methods to retrieve ice cloud properties, building ice crystal models with assumed ice crystal shape to generate the single-scattering properties is the first step. Then, with the single-scattering properties and assumed cloud properties, radiative transfer model simulations are applied to build large volume lookup tables (LUTs) that contain the relationship between cloud properties and simulated satellite sensor signals. Finally, the cloud property is retrieved by searching the best-fit cloud simulated signal to measurements. However, this conventional method requires a tremendous amount of computations and needs to improve accuracy because an optimal ice crystal model is not determined. In this study, we apply machine learning techniques to retrieve ice cloud property that attempts to improve the retrieval method and to investigate an optimal ice crystal model using satellite measurements.