Selective decay for the rotating shallow-water equations with a structure-preserving discretization

Rüdiger Brecht, Werner Bauer, Alexander Bihlo, François Gay-Balmaz, Scott P. MacLachlanarXiv 2021 · arXiv

In weather and climate modeling, a persistent challenge is maintaining numerical stability without sacrificing physical accuracy. To prevent simulations from becoming overwhelmed by small-scale noise, researchers typically introduce artificial dissipation. However, this often leads to a "leak" in the system's total energy, resulting in unphysical long-term behavior. A recent study by Rüdiger Brecht and colleagues addresses this paradox by applying the selective decay framework to the rotating shallow-water equations.

The Challenge of Numerical Dissipation

In two-dimensional turbulent flows, such as the Earth's atmosphere, energy tends to upscale while enstrophy (the square of vorticity) cascades toward the smallest scales. Without proper handling, enstrophy accumulates at the grid scale, creating numerical artifacts. Standard solutions, like biharmonic dissipation, remove this noise but also drain energy. The "selective decay hypothesis" suggests a more surgical approach: dissipate enstrophy to maintain stability while strictly preserving the system's total energy.

Methodology: Structure-Preserving Discretization

The authors bridge geometric mechanics and numerical weather prediction through three core pillars:

  1. Variational Discretization: Instead of discretizing the equations of motion directly, the scheme discretizes the underlying variational principle (Hamilton’s principle). This ensures the resulting model respects fundamental physical symmetries and conservation laws.
  2. Casimir Dissipation: By modifying the "Lie-Poisson bracket" of the system, the researchers created a framework that specifically targets "Casimir" quantities—in this case, potential enstrophy—for dissipation while leaving the Hamiltonian (total energy) untouched.
  3. Discrete Commutator on Simplicial Meshes: A key technical hurdle was implementing these complex operations on the triangular grids common in global models. The authors derived a "discrete commutator" that allows for accurate Lie bracket calculations on irregular meshes.
Geometric discretization on a triangular mesh showing the relationship between primal triangles ($T_i, T_j$), vertices ($\zeta$), and dual edges ($\tilde{e}$).

Benchmarking and Results

The method was evaluated using three classic fluid dynamics scenarios: vortex interactions, shear flow on a plane, and flow over an isolated mountain on a sphere.

Energy Conservation and Spectral Accuracy

The results demonstrate a significant leap in performance. In shear flow tests, the selective decay scheme lost approximately 100 times less energy than standard biharmonic dissipation. Furthermore, the method accurately captured the theoretical $k^{-3}$ and $k^{-1}$ power laws in energy and enstrophy spectra, respectively. Unlike blunt dissipation methods that blur fine-scale features, this "surgical" approach allows small-scale structures to remain coherent and physically meaningful.

Long-term Stability

For climate-scale modeling, long-term integration is vital. The simulations remained stable and free of grid-scale noise for integrations lasting up to 100 days, proving the robustness of the geometric approach on both flat and spherical domains.

Future Directions

While the spatial discretization perfectly conserves energy, the full time-stepping process introduces minor fluctuations. Future research aims to develop a fully energy-conserving time integrator. Additionally, extending this 2D framework to 3D "primitive equations" and handling complex physical boundaries, such as coastlines, are essential steps toward integrating selective decay into operational global climate models.

Summary updated

BibTeX
@misc{brecht2021selective,
  author = {Rüdiger Brecht and Werner Bauer and Alexander Bihlo and François Gay-Balmaz and Scott P. MacLachlan},
  title = {{Selective decay for the rotating shallow-water equations with a structure-preserving discretization}},
  year = {2021},
  eprint = {2107.02072},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2107.02072},
}
Figure