Lagrangian predictability of high-resolution regional models: the special case of the Gulf of Mexico

Abstract. The Lagrangian prediction skill (model ability to reproduce Lagrangian drifter trajectories) of the nowcast/forecast system developed for the Gulf of Mexico at the University of Colorado at Boulder is examined through comparison with real drifter observations. Model prediction error (MPE), singular values (SVs) and irreversible-skill time (IT) are used as quantitative measures of the examination. Divergent (poloidal) and nondivergent (toroidal) components of the circulation attractor at 50m depth are analyzed and compared with the Lagrangian drifter buoy data using the empirical orthogonal function (EOF) decomposition and the measures, respectively. Irregular (probably, chaotic) dynamics of the circulation attractor reproduced by the nowcast/forecast system is analyzed through Lyapunov dimension, global entropies, toroidal and poloidal kinetic energies. The results allow assuming exponential growth of prediction error on the attractor. On the other hand, the q-th moment of MPE grows by the power law with exponent of 3q/4. The probability density function (PDF) of MPE has a symmetrical but non-Gaussian shape for both the short and long prediction times and for spatial scales ranging from 20km to 300km. The phenomenological model of MPE based on a diffusion-like equation is developed. The PDF of IT is non-symmetric with a long tail stretched towards large ITs. The power decay of the tail was faster than 2 for long prediction times.