An Intercomparison of Inverse Methods Using an Eddy-Resolving General Circulation Model

Abstract

Three inverse methods (the Bernoulli, beta-spiral, and box inverse methods) are used on mean data from an eddy-resolving oceanic general circulation model, in an attempt to reconstruct the observed mean flow field. Inversions are performed in the Gulf Stream extension, a quiet region which is relatively eddy-free, the center of the region of homogenized potential vorticity, and a near-equatorial area, together with an inversion of the flow across a transoceanic sector. Resolutions for the inversions of ?°, 1° and 2° are used. Numerical estimates of geostrophy using the wider resolutions can give top-to-bottom thermal wind shears in error by up to 1 cm s?1 in a flow change of around 8 cm s?1. Two ?scores? for the methods are created, one which tests pointwise accuracy (the ?global? score) and one which tests fluxes of mass through a section (the ?flux? score). The Bernoulli method yields accurate global scores except in the homogenized region; the box inverse method yields fairly accurate global scores everywhere; and the beta-spiral only gives accurate global scores near the equator. No method gives reliable flux scores, although the box inverse was the least inaccurate, as might be expected from the nature of this method. The hypothesis of no flow at the bottom gives a predicted velocity field which is more accurate than any of the inversions most of the time. The Bernoulli and beta-spiral methods contain an internal measure which is well correlated with their accuracy, so that it is possible to estimate the accuracy of an inversion on real data.