The accuracy of cloud identification and CTH(Cloud Top Height) products from passive observation satellites often falls short. Although active observation satellites provide high-precision CTH and cloud identification information, their observational range is limited. To address these issues, a GA-LightGBM(Genetic Algorithm-Light Gradient Boosting Machine) model is proposed that utilizes data from Sentinel-5P(S5P: Sentinel-5P ), the fifth generation reanalysis data ( ERA5: Fifth generation ECMWF atmospheric reanalysis of the global climate ), and CALIPSO ( Cloud-Aerosol Lidar and Infrared Path nder Satellite Observation) to perform cloud identification and CTH prediction, respectively. The model is trained using data from June 2018 to December 2020 and tested with data from the entire year of 2021. Experimental results show that in the test set, the cloud identification model achieves an accuracy of 86% , effectively distinguishing clouds from clear skies. The cloud top height inversion model exhibits a MAE (Mean Absolute Error) of 1. 26 km, a RMSE ( Root Mean Square Error ) of 1. 87 km, and a coefficient of determination ( R2 ) of 0. 797 1,demonstrating good consistency with the true values and proving the effectiveness of the method.

cloud identification, cloud top height, lightweight gradient boosting model, ERA5, satellite remote sensing