Summary: | Parameters of a box model of the north Atlantic thermohaline circulation are optimized to fit the model results, i.e. the overturning, to data given by a more detailed climate model of intermediate complexity. Since the model is run into a steady state by a pseudo time-stepping, efficient techniques are necessary to avoid extensive recomputations and/or storing when using adjoint-based gradient representations for local optimization algorithms. The preconditioned Oneshot approach studied by Hamdi and Griewank that simultaneously updates state, adjoint and parameter values, is applied to this nonlinear climate model. For the required partial derivatives, a software tool of Algorithmic/Automatic Differentiation, namely TAF, was used. Numerical results are compared to results obtained by the BFGS-Quasi-Newton method.
|