Lagrangian predictability assessed in the East China Sea

A group of 30 surface drifters, launched over a 4 day period as part of a US Navy exercise in early October 2007, are used to assess the predictability of trajectories in a confined geographic region at the northwestern edge of the Kuroshio north of Taiwan. Model trajectories were computed from archives of hourly hindcast velocities from the US Navy East Asian Seas (EAS16) model with 1/16 degrees horizontal resolution. Three metrics are defined for comparing observed and modeled trajectories. All three metrics indicated that model trajectories separated from observations by roughly 15 km after the first 24 h on average. Because of the unique launch strategy for these drifters, with six repetitions of launches from four locations, the dependence of predictability on both launch time and launch location could be assessed separately. Predictive skill displayed only modest dependence on launch time, likely influenced by the passage of a typhoon near the experiment area a few days prior to the first drifter launch. Launch location was a much more reliable indicator of predictive skill, with trajectories for launches closest to the edge of the Kuroshio typically hardest to predict, and those for launches on the shelf, where currents tended to be weaker, predicted more accurately. Comparisons of skill metric statistics for modeled trajectories from hindcasts with and without tides suggested that tidal currents have only a small impact on predictive skill. The influence of archive time and space resolution was also studied using sets of model trajectories computed from hindcast archives that were systematically subsampled separately in space and time. Coarsening by up to a factor of eight in either space or time had little impact on predictive skill. Further coarsening degraded trajectory predictions, particularly when coarsening in time leads to an archive time step too large to adequately resolve the tides. While accurate trajectory predictions remain challenging for ocean models, skill assessments like the one presented here are important for developing error estimates for users of trajectory forecasts and for gaining new insight into potential sources of model errors. (C) 2010 Elsevier Ltd. All rights reserved.