James Test

To determine if the differences between the drifter and the respective model mean velocity fields are significant over the domain, we apply a James test (Seber, 1984), which can be used to statistically compare two vector fields with different variances. The mean velocity fields used in the test are the 2°x 2° binned velocities from the drifter trajectories and the two sets of numerical trajectories. Details of the James test are found in Appendix A. The Figures below show the results of the James test using the drifter velocities and the 0.28° and 0.1° model fields, respectively. The gray shading indicates that the null hypothesis of equal means is rejected for those bins with a possibility of being incorrect 5% of the time. In other words, the means are signficantly different in these bins when a 95% significance level is used.

Abstract

Introduction

Model Descriptions

Surface Drifters

Numerical Trajectories

Eulerian Statistics

Lagrangian Statistics

Summary

Appendix A

Appendix B

Acronyms

References

Acknowledgements

Forum

Enlarge

Drifter - 0.28° Model Velocities Comparison

The regions with significantly different means are those already identified qualitatively as being erroneous in the model field. Significantly different mean flows are found in the anticyclone off Cape Hatteras, at the point where the true GS turns to the northeast to enter the NAC, at the location where the model turns to the northeast too far across the basin, and in the region where the model displays a fictious zonal current in the southeast sector of the domain.  Also in this part of the domain, the flow is fairly quiescent and motions are rambling and circuitous so it is therefore not surprising that the weak mean flows are significantly different.

 

Enlarge

Drifter - 0.1° Model Velocities Comparison

There are fewer significantly different model and drifter means than in the 0.28° case. Again these regions were identified previously: over the mid-Atlantic Bight where the 0.l° model flow is in the wrong direction and in the CC region where the model does not resolve a coherent  flow. In all, in regions of strong currents and in most quiescent flows the drifter and 0.1° POP means do not differ significantly.