A New Cloud and Haze Mask Algorithm From Radiative Transfer Simulations Coupled With Machine Learning

Mainstream satellite cloud masking algorithms are prone to mis-masking in haze-polluted areas, which may cause errors in aerosol radiative effect calculations and attribution of surface solar radiance changes; thereby, distinguishing between clouds and haze is critical to obtaining accurate land and atmospheric data products. Existing cloud and haze mask algorithms based on the threshold method may require us to spend a lot of manpower to perform multiple threshold tests; in addition, the obtained thresholds are only applicable to particular sensors, which limits the generality of the threshold-based cloud and haze mask algorithms. In this study, a new cloud and haze mask algorithm based on a combination of radiative transfer simulations and machine learning text simulation-based cloud and haze masking (SCHM) is proposed and applied to MODIS images. When we simulated the apparent reflectance of the first seven visible and text near-infrared channels of MODIS, the CALIOP and AERONET data verification results showed that the SCHM algorithm achieved 85.16% and 90.08% hit rates for cloud and haze recognition, respectively. When we added three thermal infrared channels (20, 31, and 35 bands) for simulation, the cloud and haze hit rates were improved to approximately 85.72% and 90.62%, respectively. This indicates that the SCHM algorithm can improve the accuracy of detection results by improving the radiative transfer simulation parameters. Compared with existing threshold-based methods, the SCHM algorithm has the advantages of simple logic, convenient modification, and flexible configuration.