The GOFLOW model is a simple U-Net architecture that reads in three consecutive SST snapshots one hour apart and predicts the velocity field of the middle time step. It is trained on a joint local point velocity loss and a log spectral loss on data from the LLC4320 ocean model run. The two datasets here correspond to the predictions and ground truth fields from the held-out test set (testset_llc4320_0.2.nc) and the data and predictions from snapshots of the geostationary satellite SST product (preds_lgt_0.2_GOES_2023.nc) and include, vorticity, divergence and strain predictions. The data is in NETCDF-4 format.