Performance diagnostics for probabilistic Lagrangian drift prediction

Trajectory forecasting based on geophysical models is a useful tool for contingency and emergency aid at sea. Uncertainties in the used geophysical models, which propagate into the trajectory forecast, can be addressed through ensemble modelling. Here, we evaluate the performance of an operational ensemble prediction system on short-term forecasts for 17 undrogued drifters deployed in the Barents Sea and Fram Strait in 2022. Predicted and observed trajectories were compared by rotary spectra analysis, rank histogram, reliability diagram, and error/spread to determine the ability of the model to reproduce the observed physical processes and their uncertainties. We found that the physical processes dominating the observed spectra at inertial and subinertial frequencies are accounted for in the modelled trajectories, but that the model underestimates the energy content for higher frequencies (> 0.083 cph) with up to two orders of magnitude. Ensemble underdispersion is linked to model error rather than initial and boundary conditions. Reliability is achieved if the main forcings are accurately reproduced by the geophysical models. For highly dynamic regions, such as the Fram Strait, transient small-scale phenomena representation is critical at the uppermost ocean layer for accurate trajectory forecasting.