Adaptive mesh refinement for finite-difference ocean models: First experiments
Résumé
The application of an adaptive mesh refinement (AMR) method for structured mesh is examined in the context of ocean modeling. This method can be used with existing finite-difference ocean models at little computational and programming cost. Some first experiments in academic cases are presented in order to give some insight to the two following questions: (i) Is the AMR method appropriate and efficient for integration of numerical ocean circulation models (and particularly for long-term integration)? (ii) Can the AMR method be an efficient alternative to the classical zoom techniques for local prediction?
Numerical simulations are performed in the well-known case of the barotropic modon and in the case of a multilayered quasigeostrophic box model. They demonstrate that the use of the AMR method results in a very significant gain in CPU time (by a factor of 3) while conserving, within a 10%–20% range, the main statistical features of the solution obtained with a uniformly high resolution. For the problem of local prediction, it appears that only one simulation with the AMR method leads to better local predictions than classical nested grid techniques, wherever the region of interest is located, and for a comparable amount of computation. Further investigations are presently under way to generalize this application to basin-scale primitive equation models in realistic configurations and investigate whether or not these results are still valid.
Numerical simulations are performed in the well-known case of the barotropic modon and in the case of a multilayered quasigeostrophic box model. They demonstrate that the use of the AMR method results in a very significant gain in CPU time (by a factor of 3) while conserving, within a 10%–20% range, the main statistical features of the solution obtained with a uniformly high resolution. For the problem of local prediction, it appears that only one simulation with the AMR method leads to better local predictions than classical nested grid techniques, wherever the region of interest is located, and for a comparable amount of computation. Further investigations are presently under way to generalize this application to basin-scale primitive equation models in realistic configurations and investigate whether or not these results are still valid.