Assessing the accuracy of drift prediction products at the Naval Oceanographic Office

Customers of the Naval Oceanographic Office (NAVOCEANO) often require a prediction of object drift, either in a real-time or historical context. NAVOCEANO maintains a suite of forecasting ocean models to support real-time requests and has many years of archived model runs to support historical analysis. The spatial distribution of drift probability is a powerful tool to provide actionable information to our customers, especially when there is a significant amount of variability or uncertainty in the Eulerian velocity. This distribution is particularly useful when developing multi-year historical products of drift expectation for planning purposes or when providing real-time analysis using ensemble model runs. As with any drift product, the accuracy is only as good as the underlying model currents and algorithm used for drift. Drift probability distributions also provide a natural error metric to quantify drift forecast accuracy. Leveraging the Grand Lagrangian Deployment drifter dataset, an effort is made to make some quantification of the skill of our operational model’s drift prediction capability. Comparison of Lagrangian statistics is made between observed and modeled trajectories. Additions of different parameterizations (sub-mesoscale, windage, drag, and Stoke’s drift) are evaluated in order to see if model skill can be significantly enhanced; the effect of model resolution on prediction skill is also investigated.