AOT-560 from C2RCC

Optical Toolbox
#1

Hi,

Does C2RCC output AOT? I want to compare with other Atmospheric Correction-derived AOTS.

Regards
asim

#2

The C2RCC derives the water leaving reflectance directly from the L1b TOA reflectance, without an explicit calculation of the AOT (single step approach). The AOT is of course an important parameter included in the whole process of RT simulations and inversion, but it is implicitly handled and thus it is not an output of the C2RCC atmospheric correction.

B.t.w. if you intend to assess the quality of a water AC and compare different approach, I would recommend to compare the results of the AC, i.e. the water reflectances. This can be nicely done using, e.g. the Aeronet-OC measurement network.

You might know it already, but an interesting paper is:
ACIX-Aqua: A global assessment of atmospheric correction methods for Landsat-8 and Sentinel-2 over lakes, rivers, and coastal waters - ScienceDirect
and the related webpage ACIX II Aqua - CalValPortal

#3

Thanks for replying!

I was wondering what could be the possible reason for underestimation by C2RCC (for Rrs > 0.017 (sr-1))? Can you please suggest any read?

Best regards

#4

Could you be more specific? Underestimation in which bands? Which sensor do you use?

#5

Hi,

Sure, I have applied C2RCC in default settings on Sentimel-2 MSI images and the area of investigation is Barents Sea. It underestimate Rrs values above 0.017 (approximately) in the blue bands. The underestimation decreases from blue to red band. The pixel resolution is 60 m.

Regards
Asim

#6

C2RCC is designed for C2 waters. I understand Barents Sea is a quite complex area and it could be classified as such. The basic method of C2RCC is to “search” for similarities in the set of spectra in the set of millions of simulations which were used to train the neural net, and to interpolate in a non-linear way. We are not aware of a publication which investigated the behaviour of the neural net under various conditions, unfortunately, so any test on areas that are different from the training sets are of big interest. Maybe we can take a look together to your results and see if perharps more complex NN are needed, or what can be the issue in the blue bands.