TESTING MODEL SIMULATIONS OF THE THERMOCLINE DEPTH VARIABILITY IN THE TROPICAL ATLANTIC FROM 1982 THROUGH 1984

The multivariate model testing procedure of Frankignoul et al. has been extended to the general time-series case, thus allowing to test ocean model ability at simulating the interannual variability. The method aims at distinguishing between model inadequacies and data uncertainties; model performances are assessed from a misfit evaluated in a space of strongly reduced dimension with basis vectors issued from a double application of common principal component analysis. The testing procedure has been used to investigate the ability of Cane's linear multimode model at simulating the evolution of the 20-degrees-C isotherm depth in the equatorial Atlantic during the 1982-1984 FOCAL/SEQUAL experiment. Using Monte Carlo techniques and five different drag laws, 25 equally plausible wind-stress fields were constructed to represent the wind-stress uncertainties consistently with the sample means and variances of the original ship measurements. Even during this well-sampled period, the forcing uncertainties were substantial, with corresponding model response uncertainties as large as the interannual variability; the largest source of uncertainty is the drag coefficient indeterminacy, except in poorly sampled areas where sampling and measurement errors become comparable. Although the linear multimode model successfully simulates many features of the thermocline depth variability, there are some discrepancies with the observations, as in the Gulf of Guinea where the model poorly reproduces the eastward progression of the equatorial upwelling during summer. The multivariate analysis shows that the model-reality differences, too large to be explained by forcing and initial conditions uncertainties, are mostly due to model deficiencies. As the FOCAL/SEQUAL data provide a very stringent test of model performances, they are particularly useful for model tuning and intercomparison. The superiority of the three-mode version of the linear model over the two-mode one is thus more clearly established than in a previous comparison with the mean seasonal variations of the surface dynamic topography, and the LODYC general circulation model is shown to represent the 1982-84 changes in thermocline depth significantly better than the linear model.