The C2X nets are not available for OLCI. At the time of the C2X project, the training range for the “normal” neural nets was rather restricted to moderately turbid and absorbing coastal waters. With C2X really extreme situations were included, at the costs of less good performance in clearer waters. After this exercise, the ranges used for training neural nets for C2RCC, in general, has been extended – not as extreme as in C2X but sufficiently to cover many situations in coastal and inland waters, without scrutinizing clear waters. Thus, currently we don’t work on a version of the C2X nets for OLCI but could consider this is there is demand in the community.
For OLCI we introduced the nomenclature with Version 1 nets (V1) which are those currently used in the ground segment as well as in C2RCC delivered with SNAP6, and Version 2 nets (V2) delivered with C2RCC in SNAP 7 and used in the currently ongoing reprocessing of OLCI by ESA. Before ESA decided to accept V2 nets for the operational processing and in SNAP, these nets had the working name “alternative nets”. I.e. the alternative nets referred to in the paper are identical with V2 neural net.
@abruescas, if the version 2 nets are the ones that are described in Brockmann et al, 2016, what are then the original C2RCC nets?
I find, here is quite a bit confusion and lack of documentation.
Its difficult to understand what version uses which nets and which publication could be used as a reference.
C2RCC v1 is the one that is using EUMETSAT in the current processing of the data (OLCI L2). They use the BPAC as basis, and they offer as an alternative the C2RCC v1, but only the water quality parameters are saved and written in the output (ADG443_NN, CHL_NN, TSM_NN). The next reprocessing they will adapt the C2RCC v2.
We are actually in the process of writing the documentation.