Predictability of the Lagrangian Motion in the Upper Ocean

Piterbarg, L.I.1, Griffa, A.2,3, Mariano, A.J.2, Ozgokmen, T.M.2, Ryan, E.H.2

1 CAMS, University of Southern California, 1042 W.36th Place, DRB 155, Los Angeles, CA 90089-1113 United States
2 RSMAS/MPO, University of Miami, 4600 Rickenbacker Causeway, Miami, FL 33149-1098 United States
3 IOF, Consiglio Nazionale Ricerche, La Spezia, Italy

Abstract:

The complex non-linear dynamics of the upper ocean leads to chaotic behavior of drifter trajectories in the ocean. Our study is focused on estimating the predictability limit for the position of an individual Lagrangian particle or a particle cluster based on the knowledge of mean currents and observations of nearby particles (predictors). The Lagrangian prediction problem, besides being a fundamental scientific problem, is also of great importance for practical applications such as search and rescue operations and for modeling the spread of fish larvae. A stochastic multi-particle model for the Lagrangian motion has been rigorously formulated and is a generalization of the well known "random flight" model for a single particle. Our model is mathematically consistent and includes a few easily interpreted parameters, such as the Lagrangian velocity decorrelation time scale, the turbulent velocity variance, and the velocity decorrelation radius, that can be estimated from data. The top Lyapunov exponent for an isotropic version of the model is explicitly expressed as a function of these parameters enabling us to approximate the predictability limit to first order. Lagrangian prediction errors for two new prediction algorithms are evaluated against simple algorithms and each other and are used to test the predictability limits of the stochastic model for isotropic turbulence. The first algorithm is based on a Kalman filter and uses the developed stochastic model. Its implementation for drifter clusters in both the Tropical Pacific and Adriatic Sea, showed good prediction skill over a period of 1-2 weeks. The prediction error is primarily a function of the data density, defined as the number of predictors within a velocity decorrelation spatial scale from the particle to be predicted. The second algorithm is model independent and is based on spatial regression considerations. Preliminary results, based on simulated, as well as, real data, indicate that it performs better than the Kalman-based algorithm in strong shear flows. An important component of our research is the optimal predictor location problem; Where should floats be launched in order to minimize the Lagrangian prediction error? Preliminary Lagrangian sampling results for different flow scenarios will be presented.

Reference:

Predictability of the Lagrangian Motion in the Upper Ocean. L.I. Piterbarg, A. Griffa, A.J. Mariano, T.M. Özgökmen, E.H. Ryan, 2001 Fall AGU Meeting, 10-14 December, 2001, (NG42A-0407). Eos Trans. AGU, 82 (47),
Fall Meet. Suppl., Abstract NG42A-0407.

Links:
NG42A-0407