The global monthly mosaic generation pipeline currently ranks and selects imagery scenes based on data validity metrics derived from the Usable Data Mask (UDM2.1). While cloud pixels are properly identified as invalid and masked, snow pixels are treated as valid surface data.
In equatorial regions—such as Singapore, where snow is a physical impossibility—the UDM2.1 classifier systematically misclassifies highly reflective, optically thick cloud patches as snow. Because the mosaic scene-ranking algorithm treats snow as valid, a cloudy scene containing these misclassified "snow" patches can mistakenly outrank a cleaner, less-cloudy alternative scene. This results in corrupted, saturated surface reflectance data permanently propagating into the final monthly composite product with no warning flags for downstream consumers.
Proposed Solution
Implement an algorithmic check directly within the mosaic generation and scene-ranking pipeline to account for regional geographic realities without requiring a full retraining of the underlying UDM2.1 core classifier:
-
Latitude-Gated Rule: When processing imagery or generating mosaics for Areas of Interest (AOIs) within the tropical belt, the scene-ranking pipeline must treat any pixel flagged as suspect or cloud-equivalent.
-
Scene Re-ranking: Scenes with high "snow" percentages in these restricted tropical zones should have their quality scores penalized identically to cloud cover. This ensures the pipeline selects a truly clear alternative scene during compositing rather than a cloudy scene mislabeled as snow.
