Suspiciously high NDVI values for S2 L2A products

s2tbx
#1

Dear All,

I have been recently experiencing a hard time with NDVI values derived for Sentinel-2 scenes Level 2A (processed with Stand-alone Sen2Cor 2.4.0 with topographic correction).
My concern is that NDVI is a way too high than what I would expect for grassland/pasture land cover classes in winter/early-spring in Europe NW. Some areas are getting NDVI >0.9. The image looks ok, but the range is shifted towards the high end. I have been experiencing it for a several Sentinel-2 tiles acquired over Europe.
I am aware of this paper: http://ieeexplore.ieee.org/document/6449317/ and understand that the NDVI values derived from Sentinel-2 are going to be different, but having most of pixels in a high end on the NDVI range (much into the usual saturation levels) hampers imho for example LAI~NDVI modelling .
On the other hand, going through Sentinel-2 related scientific papers, I have never come across any publication that would have tackle this issue. On the contrary, all presented NDVI values sit well within the expected range, with only single observations on the high end (above 0.8).
I am wondering whether anyone has been experiencing similar issues with S2 data (?!).
Thanks for your input and insight.

Calculating EVI from Sentinel-2 Sen2Cor output product
#2

Dear Katarzina,

you are not the only one to observe that, and it will be a constant with atmospherically corrected products (and not only with Sen2cor). NDVI over dense vegetation is particularly sensitive to atmospheric correction errors in the red band.

You might be interested by this blog post which addresses that issue, suggesting a slight modification to the NDVI formula: http://www.cesbio.ups-tlse.fr/multitemp/?p=12746

Cheers,
Olivier

#3

Thank you Olivier for this link and your insight.

I find this blog entry consistent with my experiences with S2 data.
In one of my attempts I run Sen2Cor with maritime correction setup (instead of a standard rural) and NDVI was even higher that for a rural setup. This well agrees with the cesbio’s entry , because in a maritime model there is ‘more’ atmosphere, co consequently Red band reflectance is reduced even more.

1 Like