A new high-order finite-volume advection scheme on spherical Voronoi grids and a comparative study in a mimetic finite-volume moist shallow-water model

This paper proposes and evaluates a new class of high-order $k$-exact advection schemes for spherical centroidal Voronoi tessellations, demonstrating their high accuracy and grid robustness in both classical test cases and a mimetic finite-volume moist shallow-water model.

The Gist

Imagine the Earth is a giant, spinning basketball. Meteorologists need to predict the weather on this ball, but they can't just draw a grid of squares on it like they do on a flat map. If they tried, the squares would get squished and weird near the North and South Poles, causing the computer calculations to crash or give wrong answers.

To solve this, scientists use a special kind of grid made of hexagons and pentagons (like a soccer ball). This is called a Voronoi grid. It's flexible, meaning you can make the hexagons tiny and detailed over a specific mountain range (like the Andes) while keeping them large and simple over the empty ocean. This saves computer power while giving you a sharp picture where you need it most.

The Problem: The "Smudge" Effect

The paper tackles a specific problem with these soccer-ball grids: Advection.

In weather terms, advection is just the wind blowing things around. It carries heat, moisture, and pollution from one place to another.

• The Old Way: The current methods used in these models are like a child trying to copy a drawing by tracing it with a thick, fuzzy marker. They get the general shape right, but the details get blurry. This is called "numerical diffusion." Over time, a sharp cloud edge gets smeared out, and the weather forecast loses its crispness.
• The Challenge: Because the soccer-ball grid is made of irregular shapes (not perfect squares), it's very hard to build a "fine-tip marker" (a high-precision math tool) that works on every single cell without breaking.

The Solution: The "Laser-Sharp" New Scheme

The authors (Luan, Jeferson, and Pedro) invented a new mathematical recipe, which they call the OG scheme (named after the researchers Ollivier-Gooch who pioneered the idea on flat maps).

Here is how their new method works, using a simple analogy:

1. The Flat Map Trick: Imagine you have a bumpy soccer ball. To draw a straight line on it, you don't try to draw directly on the curve. Instead, you hold a flat piece of glass (a tangent plane) against the ball right where you are working. You do your math on the flat glass, then project the result back onto the ball.
2. The "Polynomial" Paintbrush: The old methods used a simple, straight-line paintbrush. The new method uses a curved, flexible paintbrush (a high-order polynomial). This allows it to trace the curve of the wind and the shape of the clouds much more accurately.
3. The "Upwind" Safety Net: The wind doesn't blow equally in all directions; it blows from somewhere to somewhere. The new scheme is "upwind-biased," meaning it looks at where the wind is coming from to decide how to paint the next step. This prevents the "smearing" and keeps the data stable, even when the grid cells are weirdly shaped.

The Test Drive

The authors put their new "laser-sharp" scheme to the test in two ways:

1. The "Gaussian Hill" Race:
They simulated a perfect, round hill of air moving across the globe.

• Result: The old methods (SG schemes) started to blur the hill as it moved. The new method (OG schemes) kept the hill sharp and round, even after traveling all the way around the world. The new method was especially good at handling the "variable resolution" grids (the ones with tiny cells over the Andes), proving it doesn't get confused by the changing cell sizes.

2. The "Moist Shallow Water" Storm:
They ran a full weather simulation that includes rain, clouds, and temperature.

• Result: Here, things got interesting. Even with their super-sharp advection tool, the final weather maps still showed some "grid imprinting" (faint patterns of the soccer ball grid showing through the clouds).
• The Twist: This wasn't the fault of the new advection tool! The problem was the engine driving the whole simulation (the TRiSK scheme). It's like having a Ferrari engine (the new advection) but putting it in a car with a rusty, wobbly chassis (the old fluid dynamics). The Ferrari runs great, but the wobbly chassis still makes the ride shaky.

The Bottom Line

• What they achieved: They successfully built a high-precision, "fine-tip marker" for weather models on soccer-ball grids. It is more accurate and robust than current methods, especially for keeping fine details in the atmosphere.
• The Catch: While their new tool is excellent at moving air and moisture, the underlying physics engine of the model still has some "wobbles" caused by the grid geometry.
• The Future: To get truly perfect weather forecasts, scientists need to upgrade both the tool that moves the air (which they just did) and the engine that calculates the wind and pressure (which is the next big challenge).

In short, the authors built a better wheel for the car, but they realized the car's frame still needs some reinforcement to drive perfectly on the bumpy road of the Earth's surface.

Technical Summary

1. Problem Statement

Global atmospheric models increasingly require high spatial resolutions and flexible grid structures to capture fine-scale phenomena. Spherical Centroidal Voronoi Tessellations (SCVTs), used in models like MPAS (Model for Prediction Across Scales), offer quasi-uniform spacing and the ability to perform local refinement (e.g., over the Andes) without modifying the numerical discretization.

However, the irregularity of SCVT grids poses significant challenges for high-order advection schemes:

• Grid Imprinting: Irregular cell geometries can cause numerical artifacts (grid imprinting) and convergence issues.
• Accuracy Limitations: Existing high-order schemes on SCVTs (specifically the SG2011 schemes by Skamarock and Gassmann) often fail to achieve their nominal order of accuracy due to mesh irregularity and the specific way they handle reconstruction (e.g., direct extension of 1D formulas to the sphere).
• Moist Physics Coupling: In moist shallow-water models, the interaction between high-order tracer transport and low-order dynamical cores (like TRiSK) can degrade overall performance, leading to spurious cloud/rain formation or grid imprinting in tracer fields.

The authors aim to develop a robust, truly high-order finite-volume advection scheme for SCVTs that maintains accuracy on both quasi-uniform and locally refined grids, and to evaluate its performance within a complex moist shallow-water framework.

2. Methodology

A. The Proposed Scheme (OG Schemes)

The authors propose a new class of schemes (denoted OG) based on the $k$-exact reconstruction approach by Ollivier-Gooch and collaborators, extended from planar unstructured meshes to the sphere.

• Local Tangent Plane Projection: Variables are projected onto a local tangent plane at each cell center.
• Mass-Conserving Reconstruction: Unlike the SG2011 schemes which enforce pointwise matching of cell averages, the OG schemes enforce that the reconstruction polynomial preserves the local mass (cell average) over the stencil neighbors. This is achieved via a weighted least-squares fit.
• Gaussian Quadrature: Fluxes across cell edges are computed using Gaussian quadrature (with $m=1$ or $m=2$ points) rather than a simple midpoint rule.
• Upwind Bias: The schemes utilize upwind-biased reconstructions to enhance stability.
• Order of Accuracy:
  • OG2: 2nd-order (linear reconstruction, 1 quadrature point).
  • OG3: 3rd-order (quadratic reconstruction, 2 quadrature points).
  • OG4: 4th-order (cubic reconstruction, 2 quadrature points).
• Flux Limiting: A Flux-Corrected Transport (FCT) limiter (Zalesak) is applied to prevent non-physical oscillations and negative values.

B. Comparative Framework

The new schemes are compared against the existing SG2011 schemes (SG2, SG3, SG4):

• SG2/SG4: Centered schemes (no upwind bias).
• SG3: Upwind-biased scheme.
• SG4: Extends 1D fourth-order formulas directly to the sphere using second-order derivatives.

C. Test Cases

The evaluation involves two main categories:

1. Pure Advection Tests: Standard benchmarks on the sphere (Nair and Lauritzen suite), including:
   • Solid-body rotation of a Gaussian hill.
   • Deformational flow with two Gaussian hills.
   • Deformational flow with two slotted cylinders (discontinuous).
   • Tests performed on both quasi-uniform and topography-based locally refined (Andes) grids.
2. Moist Shallow-Water Model:
   • Uses the Zerroukat and Allen moist shallow-water equations.
   • Dynamics (momentum, depth) are solved using the TRiSK mimetic finite-volume/difference scheme.
   • Tracers (temperature, water vapor, cloud water, rain) are advected using the OG and SG schemes.
   • Tests include Moist Steady Geostrophic Flow and Barotropic Instability.

3. Key Contributions

1. Extension of $k$-exact Reconstruction to SCVTs: Successfully adapted the Ollivier-Gooch methodology to spherical Voronoi grids, ensuring local mass conservation and achieving nominal convergence rates that match the actual convergence rates (unlike SG schemes which often degrade).
2. Robustness on Local Refinement: Demonstrated that the proposed schemes maintain high-order accuracy even on grids with extreme local refinement (e.g., over the Andes), showing low sensitivity to grid distortion.
3. Comprehensive Comparative Study: Provided a rigorous benchmark of OG vs. SG schemes in both pure advection and a coupled moist physics environment, highlighting the trade-offs between accuracy, stability, and computational cost.
4. Identification of Bottlenecks: Identified that while high-order advection improves tracer transport, the overall accuracy in moist simulations is limited by the low-order (TRiSK) discretization of the fluid depth and momentum equations.

4. Results

Pure Advection Tests

• Convergence Rates:
  • OG2 consistently achieved 2nd-order convergence with smaller errors than SG2.
  • OG3 achieved near 3rd-order convergence. SG3 achieved 3rd-order only on coarser grids, degrading to 2nd-order on finer grids.
  • OG4 achieved nearly 3.5th-order convergence. In contrast, SG4 failed to converge beyond 1st-order without flux limiting and struggled with grid distortion.
• Grid Distortion Sensitivity:
  • The OG schemes showed very little sensitivity to the locally refined Andes grid.
  • SG4 (centered) exhibited large errors and instability on refined grids without flux limiting.
  • SG3 (upwind) was more robust than SG4 but still less accurate than OG4 on refined grids.
• Discontinuous Fields: In the slotted cylinder test, OG4 provided the best preservation of shape and monotonicity among all schemes.

Moist Shallow-Water Tests

• Grid Imprinting: Even with high-order advection, grid imprinting (errors localized near pentagonal cells and the equator) persisted in tracer fields. This was attributed to the coupling with the TRiSK dynamical core, which uses a low-order discretization for fluid depth.
• Accuracy Comparison:
  • On quasi-uniform grids, OG4 often provided smaller errors for temperature and water vapor, especially at coarser resolutions.
  • On variable-resolution grids, SG3 generally yielded smaller errors than OG schemes for tracers, likely due to the additional error introduced by the wind reconstruction required for OG schemes.
  • Cloud Fields: SG schemes generally produced more accurate cloud fields than OG schemes, though OG4 became competitive at the highest resolutions.
• Barotropic Instability: All schemes captured vortex formation. OG3 produced sharper gradients, while OG4 was slightly more diffusive.

5. Significance and Conclusion

• Advancement in Numerical Methods: The paper successfully demonstrates that high-order finite-volume schemes based on mass-conserving $k$-exact reconstruction are viable and superior on spherical Voronoi grids compared to existing methods, particularly regarding convergence rates and robustness on locally refined meshes.
• Implications for Weather Modeling: The results suggest that while improving tracer advection is beneficial, the dynamical core (TRiSK) remains a bottleneck. The grid imprinting observed in moist simulations indicates that future improvements in models like MPAS and MONAN require upgrading the dynamical discretization (e.g., to a higher-order momentum solver) to match the accuracy of high-order tracers.
• Computational Cost: The OG4 scheme is slightly more expensive (approx. 10% more than SG3) due to the need for wind reconstruction and additional quadrature points, but the gain in accuracy and robustness makes it a competitive choice for high-resolution modeling.

In summary, the authors provide a robust, high-order advection framework that solves many of the convergence and distortion issues plaguing current SCVT-based models, while simultaneously highlighting the need for a holistic upgrade of the dynamical core to fully realize the benefits of high-order transport.