A new operational, ensemble-based search and rescue model for the Norwegian Sea and the North Sea is presented. The stochastic trajectory model computes the net motion of a range of search and rescue objects. A new, robust formulation for the relation between the wind and the motion of the drifting object (termed the leeway of the object) is employed. Empirically derived coefficients for 63 categories of search objects compiled by the US Coast Guard are ingested to estimate the leeway of the drifting objects. A Monte Carlo technique is employed to generate an ensemble that accounts for the uncertainties in forcing fields (wind and current), leeway drift properties, and the initial position of the search object. The ensemble yields an estimate of the time-evolving probability density function of the location of the search object, and its envelope defines the search area. Forcing fields from the operational oceanic and atmospheric forecast system of The Norwegian Meteorological Institute are used as input to the trajectory model. This allows for the first time high-resolution wind and current fields to be used to forecast search areas up to 60 hours into the future. A limited set of field exercises show good agreement between model trajectories, search areas, and observed trajectories for liferafts and other search objects. Comparison with older methods shows that search areas expand much more slowly using the new ensemble method with high resolution forcing fields and the new leeway formulation. It is found that going to higher-order stochastic trajectory models will not significantly improve the forecast skill and the rate of expansion of search areas.

No abstract

[1] An important aspect of particle trajectory modeling in the ocean is the assessment of the uncertainty in the final particle position. Monte Carlo particle trajectory simulations using surface currents derived from standard-range and long-range CODAR HF radar systems were performed using random-walk and random-flight models of the unresolved velocities. Velocity statistics for these models were derived from the covariance functions of differences between CODAR and drifter estimates of surface currents. Comparison of predicted trajectories and drifter tracks demonstrate that these predictions are superior to assuming the drifters stay at their initial position. Vertical shear between the effective depth of long-range CODAR measurements ($2.4 m) and that of drifters (0.65 m) causes the drifters to move more rapidly downwind than predicted. This bias is absent when standard-range CODAR currents (effective depth $0.5 m) are used, implying that drifter leeway is not the cause of the bias. Particle trajectories were computed using CODAR data and the random-flight model for 24-hour intervals using a Monte Carlo approach to determine the 95% confidence interval of position predictions. Between 80% and 90% of real drifters were located within the predicted confidence interval, in reasonable agreement with the expected 95% success rate. In contrast, predictions using the random-walk approach proved inconsistent with observations unless the diffusion coefficient was increased to approximately the random-flight value. The consistency of the random-flight uncertainty estimates and drifter data supports the use of our methodology for estimating model parameters from drifter-CODAR velocity differences.