Cloud Detection Issue in PROBA-V Collection 1 data
The modified PROBA-V cloud detection algorithm, which is used in the Collection 1 data, uses monthly land cover information, based on ESA Land Cover CCI data. Each pixel is assigned one of the following classes: ‘land’, ‘water’, ‘snow/ice’, ‘shadow’, ‘cloud’, and ‘unknown’. During the final Collection 1 validation we discovered that during the northern hemisphere winter season the amount of ‘unknown’ land cover assignments is unusually high, which eventually results in a substantial amount of clouds that are labelled ‘clear’.
This cloud detection issue impacts both the segment data (Level 2A), as well as the synthesis files at all product resolutions and is most prominent during the northern hemisphere winter season. The underdetection of clouds occurs primarily along a latitudinal band that shifts southward and reaches ~50oN as southernmost position.
We are currently investigating the cause and correction of this issue, further details will be communicated here once available.