Impact of assimilating high frequency radar surface currents on the fidelity of a Middle Atlantic Bight circulation model

The systematic establishment of coastal ocean observation systems adopting cutting-edge technology, such as High Frequency (HF) radar, satellite sensing and gliders, has provided unprecedented opportunities for improving and assessing ocean model hindcasts and forecasts. We assess the impact of assimilating high quality HF radar observations covering the Middle Atlantic Bight (MAB) into an operational New York Harbor Observation and Prediction System (NYHOPS). A nudging or Newtonian damping scheme is developed to assimilate the HF radar surface currents during operational NYHOPS observation-based spin-up (-24 to 0 forecast h). The effectiveness of data assimilation (DA) is evaluated by comparisons to surface currents derived from Lagrangian drifters tracked during a total 67 days in 2010 and 2011. The impact of DA on NYHOPS´s ability to simulate surface currents correctly during both the NYHOPS hindcast (-24 to 0 h) and free forecast (0 to 24 h) periods is analyzed quantitatively. In the MAB, DA decreased the root-mean-square-difference (RMSD) between the NYHOPS hindcast and the HF radar surface current components by approximately 3 cm s^-1 between Jun 9^th, 2011 and Jul 21^st, 2011. Compared to historic current-meter-data estimates of the dominant semidiurnal tidal current constituent, harmonic analysis of the NYHOPS surface currents after DA shows an average 1.0 cm s^-1 improvement in the semimajor axis, a 0.2 cm s^-1 improvement in the semi-minor axis, a 10° decrease in the averaged phase, and a 13° decrease in the average ellipse orientation estimates. The average RMSD between NYHOPS hindcast surface currents and those derived along the observed drifter trajectories was 13.6 cm s^-1, which, after DA, decreased by 1.1 cm s^-1 (8%). The benefit of DA during hindcast/spinup extended to the free forecast period: a decrease of approximately 5% from 21 cm s^-1 RMSD withou- DA. The average mean-vector-velocity-difference (MVVD) between surface currents from the NYHOPS hindcast with DA and currents derived along drifter trajectories was 3.6 cm s^-1, a 1.5 cm s^-1 decrease compared with that of NYHOPS without DA. Again, the benefit extended to the free forecast period, with the MVVD error decreasing from 5 cm s^-1 to 4 cm s^-1. Finally, averaged over a set of 43 particle tracking runs, the separation distance between numerical particle trajectories based on the NYHOPS currents versus observed drifter trajectories also decreased after DA (by 7% per five hindcast days and 10% per one forecast day). These results indicate that DA of HF radar currents onto operational forecast models can improve the skill of such models and, in extension, their benefit to society through, for example, more accurate search and rescue path planning.