Pixels with outliers

[andressa.muraro]

Hello,

I am facing problems with the atmospheric correction from Acolite in images of the Sentinel-2 mission.
When I compute spatial indices like NDVI, for example, many pixels are out of the “normal”. I'm using the spectral indices to highlight the water plants of a river.
(Look at the image below with the highlighted pixels)

NDVI with problem in some pixels

Capturar.JPG (21.31 KiB) Viewed 5075 times

Even in the RGB image of the acolite it is possible to see this effect in the pixels compared to the RGB image corrected by Sen2Cor:
(attach)

rgb correction from acolite

ACOLITE.JPG (17.59 KiB) Viewed 5075 times

rgb correction from SEN2COR

SEN2COR.JPG (15.56 KiB) Viewed 5075 times

Is there any way to mitigate this effect?
Do you think that the correction of the acolite is appropriate for detection of aquatic plants?

[quinten]

Hi Andressa

These are I believe the effect of the glint correction, and the threshold on the 1.6 micron channel to determine the pixels for which the correction is needed (default glint_mask_rhos_threshold=0.05).

By increasing the mask threshold (e.g. to glint_mask_rhos_threshold=0.07) the effect on the RGB is removed (actually moved closer to the shoreline). You should be a little careful as here the SWIR will also be impacted by land adjacency effects.

No glint correction:

S2A_MSI_2018_10_20_13_32_21_T22KDC_rgb_rhos_noglint.png (203.61 KiB) Viewed 5070 times

Glint correction with default threshold:

S2A_MSI_2018_10_20_13_32_21_T22KDC_rgb_rhos_default.png (199.78 KiB) Viewed 5070 times

Glint correction with increased threshold:

S2A_MSI_2018_10_20_13_32_21_T22KDC_rgb_rhos_threshold.png (199.41 KiB) Viewed 5070 times

I think the ACOLITE/DSF correction is suited for the detection of floating plants. There may be some problems caused by the interaction of the glint correction with the non-zero SWIR signal of the plants. The glint correction may be triggered depending on the magnitude of the plant SWIR signal - i.e. if it is less than the glint_mask_rhos_threshold. The SWIR signal magnitude also depends on the wetness and the pixel coverage of the plants.

There are two indices already implemented in ACOLITE that may be of use to you: the FAI (Hu et al. 2009) and the updated FAIT (Dogliotti et al. 2018).

Hope this helps!

Quinten

[andressa.muraro]

Thank you Quinten! This is really helpfull