Evaluation of GCOM-C/SGLI Cloud Flag Product with Spaceborne Cloud Radar and Lidar in Northern Hemisphere High latitudes

The cloud flag (CLFG) product from the second-generation global imager onboard the Global Change Observation Mission-Climate (GCOM-C) provides cloud/clear discrimination. It is generally difficult to distinguish clouds over surfaces covered with snow/ice. CLFG was updated in 2021 (Ver. 3) through the addition of a deep neural network (DNN) method to the previous Ver. 2 algorithm. We evaluated both versions against collocated CloudSat radar and CALIPSO lidar for October 2018-June 2019 at high latitudes in the Northern Hemisphere (over 600,000 matchup pairs, around 65–75 degrees North latitude). The results show that CLFG Ver. 3 achieved a cloud detection accuracy of 72.4%, a 2.0% improvement over Ver. 2. This improvement is attributed to the significant improvement of 8.1% for Land/Daytime scenes, where the DNN method successfully detects clouds over snow- or ice-covered surfaces which are missed by Ver. 2 algorithm. Limiting the samples to snow-/ice- covered surfaces yields an even greater improvement of 9.0% for Land/Day scenes and a net 3.1% increase across all scenes. Challenges remain in cloud detection at night when only thermal infrared channels are available (3.9% and 8.2% decrease in Overall Accuracy comparing daytime for Land and Ocean respectively). Cloud discrimination over Greenland is particularly challenging compared with other land regions, because the radiative contrast between clouds and the high-albedo surface is low. This study also examined the uncertainty information embedded in CLFG which enables user-tailored data extraction. We found that users can extract pixels with higher reliability in Ver. 3 by selecting only the most certain classes.