Some sort of clouds in water

[philbl]

Hi Quienten and all,

I'm using ACOLITE with Sentinel-2 to do some spatio-temporal analysis of turbidity in estauries. I something got some weird kind of clouds in the images. Those "clouds" increase the reflectance in the NIR, thus increase the turbidity retrieve for the image. But the things is that it's coherent with my in situ sampling.

I wonder if this is some kind of sun glint, adjency effect, some wind/wave reflections or actual patch of more turbid water.

I put two examples of image in attachments.

Thanks,
Philippe

[quinten]

Hi Philippe

This indeed looks like sun glint. For higher latitudes this occurs during summer months, and the patterns occur because of local differences in the water surface roughness.

You can try the SWIR based glint correction integrated in ACOLITE.

With no correction:

S2A_MSI_2023_08_04_15_30_07_T20TLS_L2R_rgb_rhos_default.jpeg (457.74 KiB) Viewed 15808 times

With dsf_residual_glint_correction=True:

S2A_MSI_2023_08_04_15_30_07_T20TLS_L2R_rgb_rhos_glintcorrection.jpeg (442.21 KiB) Viewed 15808 times

You will notice here some high glint spots are not corrected, but the threshold for correction can be increased.

With dsf_residual_glint_correction=True and glint_mask_rhos_threshold=0.06:

S2A_MSI_2023_08_04_15_30_07_T20TLS_L2R_rgb_rhos_glintcorrection2.jpeg (439.13 KiB) Viewed 15808 times

Note that the accuracy of the retrievals in glint areas will be reduced, and negative reflectances may occur.

I hope this helps!

Quinten

[philbl]

Hi Quinten,

It does indeed produce better quality images. On the other hand, when I use the images with the sun glint correction, I get poorer correlations with my in situ measurements (Between NIR and in situ turbidity measurements).

There's in fact a correlation between the SWIR and my in situ turbidity measurements. What I think is happening is that the correction is over-correcting the NIR band.

Does all this make sense? I've seen a few articles that use the SWIR band to extract turbidity. Would you have any advice or approaches for me to look at?

Thank you again for your quick and precise answers,

Philippe

[quinten]

Hi Philippe

Do you think the turbidity is high enough to warrant the use of the NIR band for the retrieval? Looking at the spectra derived from Sentinel-2 it seems that in addition to glint effects there is also a vegetation red-edge signal in the river, which originates from land adjacency. I will take a look at our RAdCor adjacency correction currently under development and check if this would improve the results for your river.

The SWIR is a wide spectral region, and the articles you have seen may have considered the use of a 1020 nm band for turbidity retrieval in extremely turbid waters. For the Sentinel-2 and Landsat SWIR bands at around 1600 and 2200 nm there will be no signal from below the water surface, even at very high turbidities.

A signal will start to appear in these longer wavelength SWIR bands for exposed banks and mudflats, and there the signal level will be related to the wetness of the surface. (An increase of SWIR reflectance with the mudflat drying out.) In my opinion this is no longer water turbidity.

I hope this helps!

Quinten

[philbl]

Hi Quinten,

The measured turbidity is around 0.10 g/L, with a maximum of approximately 0.2 g/L. I am using NIR because I have observed a good correlation between this band and in-situ measurements. In contrast, the correlation is not as good for the Red band, which I find somewhat surprising.

Thank you for your insights regarding the SWIR. And you're right, they used the 1240nm band

And for my knowledge, how did you spot the adjacency effect ?

Could you please let me know your findings with RAdCor? I appreciate your generosity with your time and effort. I am studying five estuaries in eastern Canada and have collected additional in-situ data. I noticed that you're looking for early adopters for your work. Maybe I could be one of them? I am currently pursuing a master's degree at INRS in Quebec, Canada.

Again, thank you for your help.

Philippe

[quinten]

Dear Philippe

My apologies for the late response. In this case, the land/vegetation adjacency effect can be spotted especially in the NIR bands of MSI. Over water pixels in the river, you can see a bump appear in these bands which corresponds to the red-edge of vegetation.

To illustrate this I extracted spectra for these points:
P1 "Field" -64.7525 E 46.4629 N
P2 "River" -64.7497 E 46.4481 N
P3 "RiverGlint" -64.7329 E 46.4605 N

Already at TOA you can spot this bump between 704 and 865 nm for P2 and P3, which is very prominent in the vegetation around the river, e.g. in P1:

spectra_rhot.png (145.35 KiB) Viewed 540 times

After ACOLITE/DSF the NIR bump is retained for both river pixels, glinted and unglinted:

spectra_rhos.png (126.82 KiB) Viewed 540 times

After ACOLITE/DSF with glint correction, the spectra for both water pixels align better, but the NIR bump remains:

spectra_rhos_glint.png (127.04 KiB) Viewed 540 times

In the last example some of the glint magnitude for P3 may be impacted by adjacency in the SWIR leading to a slight overestimate of the glint. (Either SWIR band is set to 0 to estimate the glint signal.)

Unfortunately the RAdCor early adopter programme is closed, but hopefully the code will be available soon. We are still working on how to best integrate the glint correction within RAdCor.

I hope this helps!

Quinten

[quinten]

For completeness, here are the results for the current implementation of RAdCor. The solid lines represent the RAdCor results, while the dotted lines are the results for a homogeneous surface framework (i.e. analogous to DSF, although the aerosol optical depth estimate differs):

spectra_rhos_RAdCor.png (159.05 KiB) Viewed 539 times

In this case, the water shape in the NIR is slightly better. It does seem that the aerosol optical depth/path reflectance may be slightly overestimated, for example in P2 negatives are retrieved. This could be influenced by the presence of glint in the scene and the way we are estimating the dark targets and aerosol optical depth within RAdCor. The glint signal is still present as an offset in P3. As mentioned in the post before we are working on how to integrate glint removal in RAdCor, and this result indicates we may need to do it before the atmospheric correction proper.

By manually decreasing the AOT used in the processing the negatives are reduced, but of course glint in P3 is still present:

spectra_rhos_RAdCorF.png (159.34 KiB) Viewed 511 times

Quinten