Suspiciously low reflectance values in dark areas on Sentinel-2
When evaluating MAJA 4.2.1 against MAJA 3, we noticed that the main band reflectances in version 4 are visibly lower than they used to be.
I don't have the output of MAJA 3 on hand (yet?), but see this screenshot of S2A_MSIL2A_20190401T100031_N0207_R122_T33UVQ_20190401T105727.SAFE
and LC08_L2A_192026_20190329_20190404_01_T1
Of course, the two acquisitions are from different days and times, but it should be visible that the S2 scene is darker in the forest areas. Here are some close-ups:
L8: somewhat dark, but details are visible
S2: details are lost
Both products are displayed with the same settings, with the inputs 0-1300 stretched to the display range.
You might think that the details are simply not visible because of the dynamic range, but no, the reflectance values are really 0. Here is the 0-500 range:
Here is 0-255:
For comparison, this is the L1C product, in 0-2500:
Ignoring the lower contrast and apparent haze, the details in the forest areas are clearly visible.
For example, in this area (watch at the forest slightly left to the center of the first image):
Every pixel has a value of 0:
Compared to the L1C product (only the RGB bands, this time):
In contrast, see this area (the band order is incorrect in merged_s2_all
, but it doesn't matter):
Here bands 2..4 are 1094, 1564, 2194 in the L2A and 1675, 1634, 1807 in the L1C, which are reasonably close. While in the forest case, the input values are 1181, 985, 684 and the output values are 0, 0, 0.
I'm sure that there are subtleties involved I couldn't even imagine, but this doesn't seem right.