since a couple of years I’m using S2 Band 10 to map Cirrus and cloud with an algorithm similar to Scene Classification.
I have found anomalous values for only this band since 2022 as started. I can’t say when exactly this problem has started.
In particoular i can say that reflectance values are too high for all the surface compared to before.
Anyone can tell me what changed in this data? Any information about it?
Hello @ander
You’d need to be more specific, I’m afraid. There was a major upgrade to S2 L1C and L2A products back in late January (detailed in this News post from September 2021 Open Access Hub).
Well, you are right.
As it is reported in technical docs of Scene Classification, the clear sky threshold is 0.012 and thick cloud is 0.035. I found these two threshold very reliable.
In fact L1C data for clear sky never in band 10 has values between 1 and 35, cirrus around 100-300 while thick clouds over 1000.
I’m checking 29/12/2021 R22 T32TQP and I see similar range of values.
While if i check more recent data, even in clear sky condition, i find reflectance values around 1000, that means 0.1 reflectance and obviusly my cirrus/cloud mask is fully saturated, as it is compleatly cloudy even if it is not.
I’ve just checked R22 T32_TPQ 04/03/2022.
Hope this example could help you to understand what i’m meaning.
There is an option (Tools> Options> S2TBX >Sentinel-2 Reader> add negative radio offset)to apply the offset with the sentinel product version PB04.00. By default the offset should be actived.
Please, could you check the offset options?