Satellite remote sensing characterizes fine surface information and is widely employed in military surveys, agriculture, human activity research, and other fields. Clouds cover about 60% of the sky on Earth and can block the imaging channels of optical satellites and reduce the number and quality of available pixels in images. As a special surface landform, ice-snow covers more than 40% of the northern hemisphere in winter. Both cloud and ice-snow will greatly affect the processing and analysis of remote sensing images. The spectral characteristics of clouds and ice-snow are similar in the visible light bands, which will result in unsatisfactory cloud detection over ice-snow, and misjudgment of clouds and ice-snow. In recent years, polarization detection technology has become a rapidly developing research field globally. Two polarization-loaded atmospheric aerosols directional polarized camera (DPC) and particulate observing scanning polarimeter (POSP) are carried on the hyperspectral observation satellite [Gaofen-5(02) satellite]. The "polarization crossfire" scheme of the two polarization loads has multi-angle and multi-spectrum observation capabilities with high-precision polarization and wide-swath imaging. Polar regions are covered by ice-snow all year round, and the reflectivity of clouds and ice-snow is high in the visible light bands, which makes it difficult to detect clouds in these regions. Therefore, it is of significance to conduct cloud detection research in typical regions such as polar regions based on the Gaofen-5(02) satellite data.

We employ both DPC and POSP data to perform cloud detection. First, DPC multi-angle polarimetric observations are adopted for the apparent pressure detection in the oxygen A-band. Next, multi-angle polarimetric signal clouds are added to examine water clouds over ice-snow, improving the accuracy of water cloud detection. Then, the cirrus cloud detection over ice-snow is improved by the detection of cirrus cloud bands. Finally, by analyzing the reflection properties of water clouds, ice clouds, and ice-snow in different wavebands, the waveband for the commonly utilized NDSI normalized snow index is increased to improve the detection accuracy of ice clouds over ice-snow. The optimal threshold values for each detection criterion are determined through a large number of statistical analyses of multiple regions sampled in different months.

To verify the effectiveness of the algorithm, we apply it to cloud detection over ice-snow. A total of two sample regions including the Greenland region and the Antarctic region are selected, and they are covered with ice-snow all year round. The DPC/POSP cloud detection results are in good agreement with the spatial distribution of cloud pixels from the MOD35 product [Figs. 9(b) and 9(c)]. The number of pixels is 17589. The pixel-by-pixel comparison shows that the consistency of the two products is approximately 83.3%. Among the DPC/POSP discrimination results, 26.6% are cloudy and 73.3% are clear sky, while 29.7% are MODIS cloudy and 70.2% are clear sky. This indicates that the employed DPC/POSP data are consistent with the MODIS cloud identification results when applied to cloud detection over ice-snow. In the Antarctic region [Figs. 10(b) and 10(c)] which is covered by both datasets, the DPC/POSP cloud detection results are more consistent with the spatial distribution of cloud pixels from the MOD35 product. The number of pixels is 395991. The pixel-by-pixel comparison shows that the consistency of the two products is about 94.4%. The DPC/POSP cloud detection results include 9% cloudy and 90.9% clear sky, while the MODIS cloud results contain 13% clear sky and 86.9% clear sky. 86% of the MOD35 cloud mask results are above clear sky to verify the algorithm reliability.

We propose the algorithm of cloud detection over ice-snow based on the data characteristics of polarization loads DPC and POSP in the Gaofen-5(02) satellite. The algorithm mainly includes cloud-based cloud detection, DPC multi-angle polarization signal test, cirrus cloud band detection, and improved NDSI detection. Based on the strong detection of oxygen A-band, the multi-angle polarization signal cloud is increased to test the water cloud over ice-snow, with improved accuracy of water cloud detection. Cirrus cloud bands are employed to improve the cirrus cloud detection over ice-snow. Finally, by analyzing the reflection characteristics of water cloud, ice cloud, and ice-snow in different bands, the bands leveraged by the commonly leveraged NDSI normalized snow index are improved to increase the detection accuracy of ice clouds over ice-snow. A large number of statistical analysis of multiple samples in different months helps determine the best threshold for each test judgment of each test. Greenland and Antarctic regions covered with all year round of ice-snow are selected for the algorithm verification. The consistency between cloud detection results of the proposed design algorithm and MOD35 cloud products is 83.3% and 94.4%. This indicates that this algorithm can better detect the cloud pixel over ice-snow, verifying its effectiveness.