NGWCP - Atmospheric model dynamical core

天气预报或气候模拟模型的核心是一个“动力核心”，它在可解的尺度上求解描述大气流体动力学和热力学的方程。目前气象局用于天气和气候预测的统一模型（“新动力学”）及其计划中的后续模型（“终局”）的动力学核心使用了利用经纬度网格结构的高效和精确的方法。然而，这些方案的几个方面涉及到非本地数据通信，特别是在靠近栅极的地方。这种非本地通信预计将成为未来大规模并行计算机体系结构的严重性能和可扩展性瓶颈，因为处理器之间的通信与计算本身相比要慢得多。因此，有必要研究通信需求大大减少的准均匀网格，作为未来大气模式动力核心的潜在基础。拟议中的项目将考察一些主要的备选电网。候选网格的属性将通过理论分析和使用简化方程集的数值计算相结合来检查。其中一个要求是精确的波传播。理想的方案应该尽可能地捕获所有已分解波的频率，并且不应该支持任何非物理的“计算模式”。这将在理论上检查在理想化的情况下，计算是可处理的，并在数值上更复杂的情况下。如果找到支持计算模式的候选方案，将检查控制它们的方法。另一组基本要求与质量和能量守恒有关，并且该方案应尊重基本的几何要求，例如重力势的梯度应在流体中不产生局部旋转（涡度）。在这些要求中，还需要方案能够支持平衡流动，因为观测到大气在大尺度上接近平衡。同样，这些要求将通过理论分析和数值测试的结合来检验。最后，所有候选的替代网格都有可能在数值近似中导致网格结构在数值解中反映出持续的误差，即所谓的“网格印记”。将开发测试用例并应用于量化各种候选网格的网格印迹趋势。

At the heart of a weather prediction or climate simulation model is a `dynamical core' that solves the equations describing the fluid dynamics and thermodynamics of the atmosphere on the scales that can be resolved. The dynamical core of the current Met Office Unified Model for weather and climate prediction (the `New Dynamics') and its planned successor (`ENDGame') use highly efficient and accurate methods that exploit the structure of a longitude-latitude grid. However, several aspects of these schemes involve non-local data communication, particularly near the poles of the grid. Such non-local communication is expected to become a serious performance and scalability bottleneck on future massively parallel computer architectures, for which communication between processors is disproportionately slow compared to the calculation itself. It is therefore imperative to examine alternative, quasi-uniform, grids, with much reduced communication demands, as a potential basis for a future atmospheric model dynamical core. The proposed project will examine some of the leading candidate alternative grids. The properties of the candidate grids will be examined through a combination of theoretical analysis and numerical calculations using simplified equation sets. One requirement is for accurate wave propagation. The ideal scheme should capture the frequencies of all resolved waves as well as possible and should not support any unphysical `computational modes'. This will be examined theoretically in idealised cases where the calculation is tractable, and numerically in more complex cases. If a candidate scheme is found to support computational modes, methods for controlling them will be examined. Another set of essential requirements relate to conservation of mass and energy, and that the scheme should respect basic geometrical requirements such as the fact that the gradient of the geopotential should generate no local rotation (vorticity) in the fluid. Also among these requirements is the need for the scheme to be able to support balanced flow, since the atmosphere is observed to be close to balance on large scales. Again, these requirements will be examined through a combination of theoretical analysis and numerical testing. Finally, all of the candidate alternative grids have the potential for the numerical approximations to lead to the grid structure being reflected in persistent errors in the numerical solution, so called `grid imprinting'. Test cases will be developed and applied to quantify the grid imprinting tendencies of the various candidate grids.