strobpr
commented
3 months ago I see four cases covered under this topic, which I would recommend to differentiate at threshold level:
'NoData': there is no sensor reading available for this location, e.g. outside FOV, obstruction(?), (ideally these would be distinguished -> goal)
'Invalid': an observation was acquired by the sensor, but no valid result is obtained, e.g. below noise floor, above saturation, encoding defect (ideally these would be distinguished -> goal)
'Falsified': a valid observation was acquired but it is not representing the desired measurand, e.g. cloud/snow/flooding/shadow (ideally these would be distinguished -> goal)
'Valid': these are verifyable observations of the measurand (which actually would imply that uncertainty is provided as well, which so far is only goal), e.g. ToR, LST, γ0
The first two are sensor dependent, the last two measurand dependent. The latter thus may change over the value chain,In other words a sample 'valid' e.g. in ToR, may turn 'falsified' in SR. Ideally we should provide a bit scheme for a byte binary mask that allows encoding all these options using the least possible space. The smaller they are the faster these very fundamental indicators can be transmitted. Bit decoding back to simple binary masks is fast and easy.
sjskhalsa
No Data (2.2) is differently defined in the AR PFS. This should be aligned with the other optical PFS to ensure consistency across ARD data.
Matthias' proposal: Replace “(‘empty pixels’)” with “(e.g., ‘empty pixels/invalid observation/below noise floor’)” in all PFS (except AR)
The empty pixels terminology is problematic.
We will raise at LSI-VC-15 in the combined optical PFS discussion - and also note the need to harmonise this terminology with the SAR side too.