Assessing the Real-Time Lagrangian Predictability of the Operational Navy Coastal Ocean Model in the Gulf of Mexico

This study quantitatively assesses the drift predictive skill of Fleet Numerical Meteorology and Oceanography Center’s (FNMOC’s) operational ocean models which are used to support a wide range of military and civilian applications. Overall, the findings of this work support the recommendation of spatial filtering for regional-scale ocean model velocity fields used in deep-water drift applications. In conjunction with filtering, the use of a pure particle drift algorithm is suggested for short-term forecasts and a drift algorithm including a sub-grid scale, random flight, parameterization for predictions requiring extended forecast predictions. Drift prediction skill is quantified through metrics of in-cloud percentage, distance error, and cloud size, which are used to assess the impact of different drift algorithms and underlying ocean models on the drift prediction capability. Through an exploration of parameterization additions to the drift algorithm, spatial filtering of model velocity fields, and increases in model horizontal resolution, drift prediction skill is shown to be counter-balanced on the accuracy of the model\u27s dispersive characteristics along with the accuracy of the underlying model velocity field (i.e. data-constrained, predictable features). A regional scale model at a horizontal resolution typically employed by FNMOC (3-kilometers) is found to be grossly under dispersive, and derived drift predictions using a pure particle algorithm are not skillful in terms of in-cloud percentage beyond a 24-hour forecast. Parameterization additions (i.e. sub-grid scale velocity and Leeway), which enhance model dispersion, are shown to greatly improve the regional scale model\u27s ability to predict a drift cloud that encompasses an object of interest at longer forecast lengths (\u3e 24-hours) by increasing cloud size. Increasing the model’s horizontal resolution (500-meters) is likewise shown to improve in-cloud prediction performance at all forecast lengths, due to its more accurate representation of dispersion which results in much larger cloud size predictions compared to those from a regional scale model. Spatial filtering of regional scale velocity fields using a Gaussian filter removes uncertain, unpredictable features (i.e. submesocale) leaving behind a data-constrained velocity field. Even though spatial filtering suppresses dispersion further for an already under-dispersion regional scale model, filtering is shown to significantly improve drift prediction performance extending in-cloud skill farther into the forecast, reducing distance errors by 15-20%, and reducing cloud size predictions by 20-30%