Why does FU classification of SNAP thematic water processor take Sentinel-2 MSI L1C instead of L2A data

I have noticed that the FU (Forel-Ule) classification of SNAP takes in Sentinel-2(A/B) MSI L1C data to get the water color FU index, according to my limited knowledge, for water color FU indexing, it might be better to take in the BOA reflectance ,i.e. L2A product which is properly atmoshpherically corrected. What is the key point?

The answer is as simple as it is unsatisfactory: L2A data wasn’t available at the time, and nobody did the work for L2A data again.

I implemented the addition of Sentinel-2 for the FU processor. This was done in collaboration with van der Woerd and Warnend. They provided the algorithmic details (e.g. polynomials).

The Sentinel-2 support was added in July 2017 to the codebase and operational L2A data was accessible since March 2018 (pilot data since March 2017).

the FU processor you have implemented for Seitinel-2 A/B MSI L1Cdata has demonstrated very satisfying performance for analysing the water color of Yangtse River lower stream where suffered a 2-months long drought from mid-July to mid-October ,2022. FU scale number goes down as a result of the reduction of load from non-point polluting sources influxed by the river branches.

1 Like

@macro_EOM I agree with this.


I have checked the FU results for many Sentinel-2 scenes of L1c product through the years 2016~ present, the results are reasonable to reveal the seasonal /yearly trends of water body clarities. I have also noticed that the FU indices seem to have low numbers,typically between 5 and 8, somewhat beyond my expectation of 7~14. I tried a few scenes of L2A products with FU processor (resampled to 10 m), and the results are within my expectaion range. But SNAP help tells me FU processor is specifically catered for L1c product. The question is: can I directly apply the current FU processor to L2A data and the result is logically valid?

In theory the atmospherically corrected data would give better results. But the algorithm uses polynomial coefficients which are made for the TOA data. Applying these to the BOA data would give wrong results. Even for S2A and S2B TOA the coefficients are different.

But for a final answer you should contact either van der Woerd and Warnend. Your interest might also convince them to provide new coefficients for L2A data.

Thanks to Marco for you kind and professonal directions to me. I hope SNAP future release will implement this processor to L2A MSI data.