Blue-green bands error over turbid water with Sentinel-3

[patilla7]

Hello everyone,

I'm currently working on validating Sentinel-3 atmospheric corrections over various types of inland water. I've got some really high errors when validating blue-green bands only over turbid water, but this is not happening with all data (figure attached of turbid water, grey line is in-situ Rrs data, yellow line is POLYMER estimated S3 Rrs data).

POL_Turbid_water.png

I executed POLYMER v4.16.1 using the run_atm_corr script with directional normalization disabled (normalize=2) and ERA5 auxiliary data. Did anyone else experience these errors?

[sakvaka_env]

Maybe you could show the area and the name of the OLCI file you have used?

If the outlier pixels exhibit logfb and logchl values that are significantly different from those of other pixels, the optimisation probably hasn't converged properly. Polymer v4.15 provides an option to do an initial grid search which helps with convergence issues, see the `CHANGELOG.txt`.

[fsteinmetz]

Hello,
Indeed this kind of behaviour is possible in complex cases. However, these results should be flagged, and not interpreted as valid results. See section "flagging" in README.md.
As said by sakvaka_env (thanks !), the specific test case could be useful, and the experimental OLCI configuration may improve these results, see release notes from v4.15 in the changelog.
Cheers,
François

[patilla7]

Hello sakvaka, fsteinmetz, thank you for your replies.

I'm working over several inland waters in Spain and Brazil. I checked the logchl and logfb values of two of the images, taken over Albufera de Valencia, Spain (lat 39.3353 lon -0.3525):

[1] S3A_OL_1_EFR____20220222T101229_20220222T101529_20220223T135752_0179_082_179_2340_LN1_O_NT_002.SEN3

AL220222.png

[2] S3A_OL_1_EFR____20230201T095344_20230201T095644_20230202T102859_0179_095_079_2340_PS1_O_NT_003.SEN3

AL230201.png

Image [1] has a single pixel with logchla and logfb values matching the surrounding pixels, whereas image [2] has two pixels with values that differ greatly from the others. Regarding the flagging, I have set it to the recommended value for OLCI (bitmask & 1023 == 0). Is it your recommendation to return to v4.15 to improve these results?

Kind regards,
Patricia