MSI / shift in Rw spectral levels starting in 2022

Post Reply
sakvaka_env
Posts: 25
Joined: Mon Mar 07, 2022 9:54 am
company / institution: Finnish Environment Institute
Location: Helsinki, Finland

MSI / shift in Rw spectral levels starting in 2022

Post by sakvaka_env »

Hi!

We have been using Polymer 4.13 for the retrieval of Rrs spectra (especially at the green wavelength) from Sentinel-2 MSI and Sentinel-3 OLCI data over the Baltic Sea. In previous years, Rrs @ 560 nm on MSI and OLCI from Polymer have always had comparable values, but starting from 2022, we are seeing an overestimation of Rw560 on MSI with respect to OLCI on collocated scenes although nothing has changed in our processes. The difference is almost constant, about 0.01 Rrs units, and concerns both Sentinel-2A and Sentinel-2B.

I installed Polymer 4.14 to see whether the same problem occurs in the newer version, too, and it does seem to. In pre-2022 data, Rw560 are comparable on MSI and OLCI for collocated scenes, but for this year, they have diverged and MSI Rrs is systematically about 0.01 Rrs units above OLCI Rrs. Otherwise, they show similar behaviour. Atmospheric conditions shouldn't impact since we are correcting both with the same NASA ancillary data.

In the following Google Drive link, there is one case example of how the difference looks like in practice in the Northern Baltic Proper. The scenes have been processed with Polymer 4.14, then cloud masked and mosaicked into daily composites.

https://drive.google.com/drive/folders/ ... sp=sharing

At least one change that has happened to Sentinel-2 since last year has been the new ESA data format (baseline 4.00). As Polymer 4.14 was released in December last year, we were speculating if that is the root cause of the issue.

Has anyone noticed anything similar?

EDIT:

A wild guess – in the Level-1 MSI reader (level1_msi.py), line 336 doesn't seem to take into account the new band-specific constant RADIO_ADD_OFFSET, defined in the Datastrip and User Product Metadata. Starting from baseline 4.00, DNs are converted to TOA reflectances with the formula (L1C_DN + RADIO_ADD_OFFSET) / QUANTIFICATION_VALUE.
Post Reply