Smile Effect correction for post-processed Sentinel 3 SYN AOD

[Frankie91]

Hello everyone,

I’m currently trying to use the post-processed S-2 SYN AOD data produced by the following research

Deep-learning-based post-process correction of the aerosol parameters in the...

The data is generated through the application of a post-processing algorithm which combines the S-2 SYN, OL-1-ERR and SL_1_RBT files.Unfortunately, as you can see from the following image, the smile effect issue still affects the quality of the results:

I’ve already looked for a solution in the forum and found mention here:

Stripes in Sentinel-3 OLCI 761.25 (band 13) and 764.375 (band 14)

Hi there – I’m using the s3tbx to successfully transform OLCI radiances to TOA reflectances. However I notices only that there are some evident stripes in the aforementioned bands, you can see the stripes in the following screenshot which shows the radiances: [image] I checked several images and all of them have the same issue. Am I missing something? is any additional correction for these absorption bands needed? Thanks! Gerardo

and here:

OLCI smile correction

Nice thought, but no it is not. I hope that a smile correction can be released sometime early next year. We are waiting for external input. As soon as we have it it should not take too long.

that a processor was in development some years ago to correct this issue. Does anyone have information about whether or not this processor was indeed developed and added to Snap? Or if any alternative solution for this issue exists(maybe through some clever data cleaning in python?)

[marpet]

The development of the smile correction processor has been abandoned some time ago. The results were just not good enough.
I’m also not aware of an alternative.
Maybe someone else can help?

Edit: Corrected the text. Thanks @gnwiii

[gnwiii]

@marpet

[emphasis added] development of the smile correction processor has been abundant some time ago.

I assume “abundant” should be “abandoned”, the sort of typo modern software likes to sneak into documents, but potentially confusing to non-English readers and machine translators.

[Frankie91]

Thanks to both for your replies - guess I’ll just plan my work around the presence of this effect then.

if I may ask, @marpet, do you know whether any deep learning approach was tried for the processor? I have some ideas in mind, but I’d like to know if they have already been unsuccessfully attempted.

[marpet]

I’m not aware of such an approach.
But I can say that we have reactivated the work on the smile correction. Coincidentally at the same time as your question. This should not mean that it will be available next week. It still can take some months or we might stop further development again.

[Frankie91]

I see, thank you for sharing the update and good luck!

I’ll be also experimenting on how to mitigate this effect with such approaches for my thesis in the meantime, although with a more restricted focus on the aforementioned AOD data.