[A-TT30] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT30-02] Artificial Intelligence Technology to Retrieve Cloud Properties Using Geostationary Satellite Measurements

*Feng Zhang^1,2、Zhijun Zhao²、Wenwen Li¹、Jingwei Li² (1.CMA-FDU Joint Laboratory of Marine Meteorology, Department of Atmospheric and Oceanic Sciences & Institutes of Atmospheric Sciences, Fudan University, Shanghai, 200438, China、2.Key Laboratory for Information Science of Electromagnetic Waves, Ministry of Education, School of Information Science and Technology, Fudan University, Shanghai, 200438, China)

キーワード：image-based transfer learning model (ITLM), advanced geostationary radiation imager (AGRI), Fengyun-4A satellite, cloud physical parameters, Tibetan Plateau

Satellite remote sensing serves as a crucial means to acquire cloud physical parameters. However, existing official cloud products derived from the advanced geostationary radiation imager (AGRI) onboard the Fengyun-4A geostationary satellite suffer from limitations in computational precision and efficiency. In this study, an image-based transfer learning model (ITLM) was developed to realize all-day and high-precision retrieval of cloud physical parameters using AGRI thermal infrared measurements and auxiliary data. Combining the observation advantages of geostationary and polar-orbiting satellites, ITLM was pre-trained and transfer-trained with official cloud products from advanced Himawari imager (AHI) and Moderate Resolution Imaging Spectroradiometer (MODIS), respectively. Taking official MODIS products as the benchmarks, ITLM achieved an overall accuracy of 79.93% for identifying cloud phase and root mean squared errors of 1.85 km, 6.72 µm, and 12.79 for estimating cloud top height, cloud effective radius, and cloud optical thickness, outperforming the precision of official AGRI and AHI products. Compared to the PRFM (pixel-based random forest model), ITLM utilized the spatial information of clouds to significantly improve the retrieval performance and achieve more than a 6-fold increase in speed for a single full-disk retrieval. Against the active remote sensing dataset, ITLM also demonstrated stable performance during the nighttime and seasonal cycles. Moreover, the official MODIS and AHI products were found to be inadequate for characterizing the spatiotemporal distribution of clouds over the Tibetan Plateau (TP). Therefore, the AGRI ITLM products with spatiotemporal continuity and high precision were used to accurately describe the spatial distribution characteristics of cloud fractions and cloud properties over the TP during both daytime and nighttime, and for the first time provide insights into the diurnal variation of cloud cover and cloud properties for total clouds and deep convective clouds across different seasons.