How are the diffuse attenuation coefficient parameters (kd 490,kdmin,kd_z90max) retrieved in C2RCC?

I have processed S3 OLCI and S2 MSI data with C2RCC processor and checked the diffuse attenuation coefficient parameters, i.e. kd 490,kdmin,kd_z90max, for the lakes/large water bodies in my province,. the zd_z90max as a representative of water transparency showed good correration with Secchi Disk Depth of the ‘ground truth’. But very few information is available about the retrieval of the diffuse attenuation coefficient related parameters, kd 490,kdmin,kd_z90max. There are many research papers covering the remote sensing retrieval of diffuse attenuation coefficients and the relationship to SDD. Can anyone elaborate on the algorithm/fomulas of C2RCC for kd_z90max? Thanks a lot.

The kds are part of the in-water neural net, which uses the water leaving reflectances derived by the AC_NN (forward model). These Lw are taken, and an inverse model is used to generate IOPs and TOA radiances. They are an optional feature on the input water leaving reflectances, which derives the diffuse attenuation coefficient (443 nm) by a dedicated neural net.