Validation of the SatGEM VelocityDataset using Argo Data

The 4D SatGEM velocity model of the Southern Ocean has been derived from satellite altimetry and a Gravest Empirical Mode (GEM) model of temperature and salinity. The validity of the SatGEM model is assessed in this study by numerical forward modelling of Argo float trajectories through the 4D velocity fields. Univariate statistical analysis of simulated and actual path lengths, together with analysis of the misfit between simulated and actual surfacing positions, suggests that the SatGEM model provides an adequate representation of the gross features of the Southern Ocean. However, on a local scale there is significant variability in misfit due to a combination of statistical fluctuations, the limited SatGEM model resolution and complex local velocity fields. A high proportion of large relative misfit cycles are apparent for short Argo path lengths suggesting that the Argo float trajectories in these areas may be strongly influenced by complex small-scale flow features that cannot be adequately resolved in the SatGEM model. Argo float cycles with path lengths in excess of 50km and small relative misfits are effectively randomly distributed. However statistically significant clusters of large relative misfit cycles are apparent particularly in areas of known small-scale eddy current development such as east of the Kerguelen Plateau and also in the Brazil-Falklands Convergence where GEM models based on satellite altimetry are known to be inaccurate. The angular misfit for medium and small relative misfit Argo cycles is normally distributed suggesting that variability in this case arises due to statistical processes. The uniform distribution of angular misfit for large relative misfit cycles is suggestive of more random chaotic processes likely related to flow bifurcations in zones with complex velocity structure. In these cases the relative misfit is highly sensitive to the initial descent position of the float in the velocity model. Validation of the SatGEM model using Argo float data suggests that velocity models derived from satellite altimetry and GEM derived temperature and salinity fields can provide adequate representations of ocean dynamics for large-scale investigations but the small-scale applications of these models are limited primarily by their spatial resolution and hence their inability to accurately model complex local flow features.