Conclusions

The adjoint method is shown to succeed in decreasing the cost function. Such an outcome was not a priori obvious, in view of the complexity of the cost function expression and the introduction of the open-boundary conditions as control variables. The adjustments made to the control variable are physically sensible, consistent with the known deficiencies of the model, and with the results obtained on a global scale by Stammer et al. (1999). More iterations are needed however, to reduce the cost function to a level where one can claim a fully consistent solution.  
The next steps of this work include the use of a more complete model mixed-layer (KPP) and of additional data in the form of XBT and P-ALACE measurements of the temperature and salinity fields. Experiments have also been undertaken using a variable grid in latitude with a higher resolution in the tropics.