| C2RCC Algorithm Specification |
|
The C2RCC processor is based on deep learning approaches. Neural networks are trained in order to perform the inversion of spectrum for the atmospheric correction, i.e. the determination of the water leaving radiance from the top of atmosphere radiances, as well as the retrieval of inherent optical properties of the water body. The C2RCC processor relies on a large database of simulated water leaving reflectances, and related top-of- atmosphere radiances. A careful characterisation of optically complex waters through its IOPs as well as of coastal atmospheres is used to parameterise radiative transfer models for the water body and the atmosphere. Covariances between the water constituents are taken into account and a large database of reflectances at the water surface is calculated. These reflectances are further used as lower boundary conditions for the radiative transfer calculation in the atmosphere. Finally, a database of 5 million cases is generated, which is the basis for training neural nets. For example, the top-of-atmosphere full spectrum is input to a neural net, and the water leaving reflectance in the visible and near-infrared bands is the output. The training can be understood as a nonlinear multiple regression.
The input spectra are corrected for gaseous absorption. Air pressure, and thus a proper altitude correction, is inherent part of the neural network processing. The main output of the atmosphere part are directional water leaving reflectances produced by the atmospheric correction neural net. The atmosphere part contains out-of-range tests and out-of-scope tests of the TOA reflectances, resulting in corresponding quality flags. Optionally the output of the auto-associative neural net used of the out-of-scope test can be written to the output file in the SNAP version of the processor. The output from the transmittance NN is also used to raise a cloud-risk flag. The in-water part gets as input the directional water leaving reflectances from the atmosphere part.
The general concept is described in the ATBD for OLCI L2 Ocean data, but also applicable to other sensors like,
S2-MCI and Landsat8:
OLCI Level 2 Algorithm Theoretical Basis Document Ocean Colour Turbid Water
The general concept is described in ATBD for MERIS 4th reprocessing data:
Algorithm Theoretical Bases Document (ATBD) for L2 processing of MERIS data of case 2 waters, 4th reprocessing
Description of the evolution of the algorithm:
Evolution of the C2RCC neural network for SENTINEL-2 and 3 for the retrieval of ocean colour products
in normal and extreme optically complex waters