This dataset contains a selection of the CORA database (http://doi.org/10.17882/46219) for the Gulf Stream Extension region, with several gridded products colocated at the time and location of these hydrographic profiles: bathymetry, MDT, SST, SLA and its derivatives (UGOS, UGOSA, VGOS, VGOSA). The selection of CORA hydrographic profils is done for 1993-2019, down to 1000m (only profiles extending between 25 and 1000m are kept) and with QC = 1 for both temperature and salinity measurements. This dataset was used to train the neural network presented in Pauthenet et al (2022, https://doi.org/10.5194/egusphere-2022-25). The OSnet neural nework was trained to learn the link between the hydrographic profiles (target) and the gridded data (input).