Fig. 1. Spatial distribution of radiosondes after matching

Fig. 2. Spatial matching of GIIRS FOV and AGRI pixel

Fig. 3. (a) The GIIRS observed 900 cm⁻¹ brightness temperature; (b) Matched cloud coverage CFR; (c) Matched cloud optical thickness COT from 00 to 01 (UTC) on 10 July 2021

Fig. 4. U-Net model structure

Fig. 5. Radiosonde data and GIIRS retrieval in Beijing at 20:00(UTC) on 22 June 2021. (a) Temperature; (b) Water vapor maxing ratio profiles (Dashed lines are Radiosonde data, solid lines are GIIRS retrieval)

Fig. 6. Temperature retrieval error profiles of 2021. (a) Winter RMSE; (b) Winter ME; (c) Summer RMSE; (d) Summer ME (Red lines are clear-sky field of view, black lines are all-sky, blue lines are GIIRS Level 2 product; February to March in winter, June to July in summer)

Fig. 7. Temperature retrieval error profiles at different altitudes. (a) RMSE; (b) ME (Red lines are statistics for low altitude, black lines are for medium altitude, blue lines are for high altitude, and the samples are winter samples)

Fig. 8. The same as Fig.6 except for water vapor maxing ratio retrieval error profiles

Fig. 9. Temperature retrieval error profiles for different cloud coverage. (a) Winter RMSE; (b) Winter ME; (c) Summer RMSE; (d) Summer ME (Red lines are clear-sky, green lines are less cloudy, blue lines are heavily cloudy, black lines are fully cloudy)

Fig. 10. The same as Fig.9 except for water vapor maxing ratio retrieval error profiles

Fig. 11. Temperature retrieval error profiles for different cloud optical thickness. (a) Winter RMSE; (b) Winter ME; (c) Summer RMSE; (d) Summer ME (Red lines are thin clouds, blue lines are medium thickness clouds, black lines are thick clouds)

Fig. 12. The same as Fig.11 except for water vapor maxing ratio retrieval error profile