Simulated Annealing: A Tool for Data Assimilation into an Almost Steady Model State

View More View Less
  • 1 Alfred-Wegener-Institute for Polar and Marine Research, Bremerhaven, Germany
© Get Permissions
Full access

Abstract

A new method is used to calculate a steady-state best fit of a given strongly nonlinear time-dependent model to observed data. The proposed technique has a statistical nature and is known as simulated annealing. It is described in detail and two examples are presented. In the first example a simple but highly nonlinear model is considered. It is shown that simulated annealing is robust and converges to the solution, whereas the adjoint technique, a sophisticated optimization method, fails. The second example illustrates that simulated annealing is also able to handle a problem with many degrees of freedom (∼4000). The required amount of computer time can be accepted.

Abstract

A new method is used to calculate a steady-state best fit of a given strongly nonlinear time-dependent model to observed data. The proposed technique has a statistical nature and is known as simulated annealing. It is described in detail and two examples are presented. In the first example a simple but highly nonlinear model is considered. It is shown that simulated annealing is robust and converges to the solution, whereas the adjoint technique, a sophisticated optimization method, fails. The second example illustrates that simulated annealing is also able to handle a problem with many degrees of freedom (∼4000). The required amount of computer time can be accepted.

Save