The Dual-channel Dynamic Threshold Method And 6s Mode Automatic Cloud Detection And Quantitative Calculation Of The Apparent Reflectivity Of The Gms-5, And Fy2b Images

This paper presents two methods for cloud detection: multi-channel dynamical threshold algorithm and apparent reflectance algorithm. The dynamic threshold algorithm is based on a fact that measurements at the chosen channels are in different scopes between the different observation objectives (here as earth surface and cloud). In this research, a statistical histogram analysis is performed based on 32x32-pixels segments on the images. Discussions about how segment and bin sizes effect on the cloud masking show that it is appropriate to chose 32x32 as segment size,, Using automatic and dynamical threshold algorithm, cloud-contaminated pixels are masked on GMS5 images, FY2B raw images and images wiped off stray light. The cloud mask results are compared with control masks that are created by visual inspections. In low and middle latitudes, the algorithm executes with a better precision than in high latitudes. The low surface temperature, feeble sun illumination and more snow cover area in high latitudes may cause the poor cloud masking quality. The cloud masks on FY2B images wiped off stray light are better than the raw images.The apparent reflectance algorithm estimates the clear-sky visible apparent reflectance as the bases of visible channel thresholds with the 6S radiation transfer model. Cloud detection is tested with apparent reflectance algorithm to GMS5 images. The cloud detection results are also compared with control masks that are created by visual inspections. In case the multi-day clear-sky composition is contaminated by cloud cover, it adds the chance of clear score failure. In case the multi-day clear-sky composition there is no snow cover, but in the same area at the image for detection there is snow, it also adds the chance of clear score failure.Finally, the MODIS cloud detection results are used as validation for both algorithms tested with GMS5 images in this research. The two algorithms both have a global score around 80%. The validation result indicates that the clear failure score is relative higher than the cloud failure score. This implies that the two algorithms tested in this research are more strict for clear pixel diagnosis than for cloud. The validation result also indicates that over land, the apparent reflectance algorithm behaves better than dynamical threshold algorithm; while over ocean, the clear and cloud failure scores do not have much difference. This implies that the apparent reflectance algorithm improves the cloud detection over land.