A regional implementation of a mixed finite-element, semi-implicit dynamical core
Christine Johnson, Ben Shipway, Thomas Melvin, Thomas Bendall, James Kent, Ian Boutle, Alex Brown, Mohamed Zerroukat, Benjamin Buchenau, Nigel Wood · 2025
Read A regional implementation of a mixed finite-element, semi-implicit dynamical core by Christine Johnson, Ben Shipway, Thomas Melvin, Thomas Bendall, James Kent, Ian Boutle, Alex Brown, Mohamed Zerroukat, Benjamin Buchenau, Nigel Wood (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-11528.txt.
Introduction
Ten authors adapt a new weather model core so it can run over a limited region instead of the whole globe. Regional models take their edge conditions from a coarser driving model. The core, built by the Met Office with a semi-implicit time scheme and mixed finite elements, had no way to accept them. The team adds the boundary data to the pressure and momentum solve at the heart of each step. They also adjust the solver aids that make that system fast to compute.
They test the regional version with big-brother experiments, where an idealized full run supplies the truth. The nested subdomain results matched the full-domain ones, which confirms the boundary feeding works. When the supplied boundary data was imperfect, blending smoothed away the clashes. The authors present the work as the bridge that lets the new core serve regional forecasting.
Catalog record
| CREATOR | Christine Johnson, Ben Shipway, Thomas Melvin, Thomas Bendall, James Kent, Ian Boutle, Alex Brown, Mohamed Zerroukat, Benjamin Buchenau, Nigel Wood |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-11528
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A regional implementation of a mixed finite-element, semi-implicit dynamical core Christine Johnson, Ben Shipway, Thomas Melvin, Thomas Bendall, James Kent, Ian Boutle, Alex Brown, Mohamed Zerroukat, Benjamin Buchenau, Nigel Wood Abstract This paper explores how to adapt a new dynamical core to enable its use in one-way nested regional weather and climate models, where lateral boundary conditions (LBCs) are provided by a lower-resolution driving model. The dynamical core has recently been developed by the Met Office and uses an iterated-semi-implicit time discretisation and mixed finite-element spatial discretisation. The essential part of the adaptation is the addition of the LBCs to the right-hand-side of the linear system which solves for pressure and momentum simultaneously. The impacts on the associated Helmholtz preconditioner and multigrid techniques are also described. The regional version of the dynamical core is validated through big-brother experiments based on idealised dynamical core tests. These experiments demonstrate that the subdomain results are consistent with those from the full domain, confirming the correct application of LBCs. Inconsistencies arise in cases where the LBCs are not perfect, but it is shown that the application of blending can be used to overcome these problems. Keywords — regional, limited area model (LAM), dynamical core, finite element, lateral boundary conditions (LBCs), multigrid † † footnotetext: This pre-print has been submitted to the Quarterly Journal of the Royal Meteorological Society. 1 Introduction In weather and climate forecasting, improved accuracy can be obtained by running the numerical model with finer resolution grids and more sophisticated representations of the physical processes. However, this needs ever more computing power. In recent years that increase in computing power has been achieved by changes to supercomputer architectures. Over the last decade there has been more focus on parallel processing, requiring numerical models to improve their scalability. And in the near future, supercomputers will have a greater diversity of hardware, requiring bespoke optimisations and programming languages. To harness the power of these diverse supercomputer architectures, the Met Office has developed a new dynamical core, GungHo [ 1 , 2 ] . This addresses the scalability bottleneck caused by the singularity at the poles on a global longitude-latitude grid by switching to a cubed-sphere mesh. Additionally, the numerical schemes for transport have been improved using new inherently conservative and consistent finite-volume schemes [ 3 ] . This new dynamical core also uses a new infrastructure, known as LFRic [ 4 ] , which was created with a design philosophy of ‘Separation of Concerns’, separating the science code from supercomputer optimisation code and enabling use on diverse hardwares. The Met Office uses both global and regional models for weather and climate forecasting. Global models predict the evolution of large weather systems and climate signals over days to years, while regional models provide enhanced resolution and detail over smaller, limited-area regions, with the Met Office using a rotated-pole, lat-lon grid. These regional models form the basis of the Met Office’s operational high-resolution numerical weather predictions [ 5 , 6 , 7 ] , regional climate studies [ 8 ] , convective-scale ensembles [ 9 ] and more recently sub-km, hectometric scale models [ 10 ] . Therefore, a regional version of the GungHo dynamical core is essential for our future forecasting capabilities. The concept of regional models is not new, and various studies have examined how to handle lateral boundary conditions (LBCs). Early studies focused on finding appropriate inflow and outflow boundary conditions to give well-posed mathematical equations e.g. [ 11 ] . Whilst this approach has been implemented in simple models e.g. [ 12 ] , it is not straightforward for multi-level, full-models with inconsistencies in the boundary data. The alternative approach known as relaxation [ 13 ] or blending [ 14 ] is commonly used. The method, also adopted here, allows the interior solution to diverge from the driving model while ensuring that the solution matches at the boundaries by blending LBC data with the interior solution. This paper describes the limited area model version of the GungHo dynamical core. We begin by describing the approach to creating a regional model — a general philosophy applicable to generating a regional model for any dynamical core. We then describe the continuous equations and spatial and temporal discretisations that form the basis of the GungHo dynamical core and how these discretisations can be adapted for the regional model. This is followed by computational experiments based on idealized dynamical core tests. 2 Philosophy of Approach The aim here is to use the regional model in a one-way nesting, where the regional model is nested in a low-resolution driving model and information is passed from the driving model to the regional model, but not vice versa. This is in contrast to other philosophies such as mesh refinement e.g. [ 15 ] , mesh stretching e.g. [ 16 ] , and two-way nesting e.g. [ 17 ] that are used elsewhere. To enable this one-way nesting we first consider the situation where the regional model is able to exactly replicate the driving model over the limited area region. This is analogous to the big-brother and perfect-model experiments described in for example, [ 14 ] and [ 18 ] , where the solution from the subdomain is compared with the solution from running the model over the entire domain, and the LBC data is derived from the full-domain run where the regional model has exactly the same resolution and numerical schemes as the driving model. The big-brother approach can be summarised as: 1. Within the subdomain, the regional numerical model updates the values. 2. Outside the subdomain, the values are updated by copying data from a scenario where the numerical model has been run over the entire domain (with the same numerical schemes and spatial and temporal resolutions). This is the lateral boundary condition (LBC) data. 3. The results should be identical (within the tolerance of the numerical implementation) to those obtained by using the numerical model to update values over the whole domain. The big-brother experiment is typically used to verify the validity of the regional model, but in this paper we extend the concept further by using points 1-3 to design the regional model. The aim is to design the regional model by adapting the full model equations to ensure that points 1-3 are satisfied. This formulation can then be extended later, through the use of blending, to allow for the case where the regional model is different to the model used to create the LBC data, for example with a lower-resolution driving model. 3 Dynamical core for a global domain To explain how the dynamical core for the regional domain is derived, we first describe the dynamical core for the global domain. This is essentially the same as the formulation presented in [ 2 ] , but additionally using the eliminated density and potential temperature in the solver as described in [ 19 ] and the SWIFT transport scheme of [ 3 ] . 3.1 Continuous Equations Central to the dynamical core is the solution of the Euler equations for a perfect gas in a rotating frame, with the equation of state, \displaystyle\frac{\partial\mathbf{u}}{\partial t} \displaystyle=-\mathbf{u}\cdot\gradient\mathbf{u}-2\bm{\Omega}\times\mathbf{u}-\gradient\Phi-c_{p}\theta\gradient\Pi (1) \displaystyle\frac{\partial\rho}{\partial t} \displaystyle=-\gradient\cdot(\rho\mathbf{u}) \displaystyle\frac{\partial\theta}{\partial t} \displaystyle=-\mathbf{u}\cdot\gradient\theta \displaystyle\Pi^{\left(\frac{1-\kappa}{\kappa}\right)} \displaystyle=\frac{R}{p_{0}}\rho\theta where the