Thoughts Sentinel-2 banding/striping artifacts over water

[beder]

Hi all,

There is a well-known occurrence in Sentinel-2 (S2) imagery in which striping artifacts appear over water. I've read that this striping is an expected occurrence or side-effect due to the arrangement of the S2's detectors and could also depend on season, water conditions, atmospheric conditions, tile location on the swath, as well as orientation of the satellite in relation to the earth/sun (azimuth, zenith, elevation angles).

I've attached some results which these striping effects appear.

Has anyone thought to use acolite to address these artifacts? Is this possible? Alternatively, has there been any discussion about how to programmatically predict/detect such occurrences based on a set of factors/parameters?

I browsed the forum before posting this, and did not see any discussion on this topic. Apologies if it has been covered before.

Thank you.

[quinten]

Hi Bryan

The detector banding is most common in sun glint viewing conditions, i.e. eastern end of the swath, and becomes more obvious at lower latitudes, where the observations are generally closer to the principal sun glint spot. The banding effect is indeed caused by the sensor construction, alternating detectors are mounted staggered and upside down which impacts the acquisition timing and hence relative azimuth. Due to the differences in azimuth each detector measures a different part of the sun glint. The problem is made slightly worse by the fact that spectrally adjacent bands can be at opposite ends of a single detector. The azimuth impacts on the atmospheric reflectance are very small (0.5% or so?).

Using the resolved geometry and a good estimate of the wind speed you can get a reasonable correction in ACOLITE (generic beta version on GitHub). For these nadir viewing geometries, the glint modeling is however very sensitive to wind speed, and computing and subtracting glint gives very mixed results. The default wind speed source (NCEP/MET global model) is usually not accurate enough for a given scene. For best results with glint correction in the current generic ACOLITE you also have to set the DSF to fixed AOT estimation (dsf_aot_estimate=fixed).

I get the following outputs setting s2_target_res=60 and dsf_aot_estimate=fixed, which retrieves quite some negatives in the less glinted parts:

S2B_MSI_2018_06_19_15_35_55_T19TDG_L2R_rgb_rhot_small.png (274.61 KiB) Viewed 31567 times

S2B_MSI_2018_06_19_15_35_55_T19TDG_L2R_rgb_rhos_small.png (242.51 KiB) Viewed 31567 times

I have been looking at using the NIR/SWIR bands to improve the wind speed estimate from the image itself, but this code is not public or integrated in ACOLITE yet. You can also use the glint correction that estimates the glint from the residual SWIR signal. (But at the moment this is also not good enough especially in strong sun glint due to the alternating detector average and band azimuth differences.)

In direct glint viewing conditions and rather clear waters you will probably get much better results from using a method that models and fits the water reflectance, such as POLYMER.

Quinten