Abstract This study develops and tests a version of the Python-driven, non-hydrostatic Model for Prediction Across Scales – Atmosphere (MPAS-A) dynamic model, as well as its tangent linear and adjoint models. The non-linear, non-hydrostatic dynamic core of the MPAS-A is restructured to have a Python driver for the convenience of parsing namelists, manipulating matrices, controlling simulation time flows, reading model inputs, and writing outputs, while the heavy-duty mediation and model layers are retained in Fortran for computational efficiency. Under the same Python-driving structure, developed are the tangent linear and adjoint models for the dynamic core of the MPAS-A model with verified correctness. The case of Jablonowski and Williamson’s baroclinic wave is used for demonstrating the approximation accuracy of the MPAS-A tangent linear model and the applicability of the MPAS-A adjoint model to relative sensitivity studies. Numerical experimental results show that the tangent linear model can well approximate the temporal evolutions of non-linear model perturbations for all model variables over a four-day forecast period. Employing the MPAS-A adjoint model, it is shown that the most sensitive regions of the 24-h forecast of surface pressure are weather dependent. An interesting westward vertical tilting is also found in the relative sensitivity results of a 24-h forecast of surface pressure at a point located within a trough to model initial conditions. This functionality of the MPAS-A adjoint model is highly essential in understanding dynamics and variational data assimilation. Plain Language Summary The MPAS-A is an advanced global numerical weather prediction model with a hexagonal mesh that can be compressed for higher resolutions in some targeted regions of interest and smoothly transitioned to coarse resolutions in others. In this study, a Python-driven MPAS-A model is first developed, combining a flexible Python driver and Fortran’s fast computation, making the MPAS-A model exceedingly user- and platform-friendly. The tangent linear and adjoint models of the MPAS-A dynamical core are then developed, both of which are required for various sensitivity studies. They are also indispensable components of a future MPAS-based global four-dimensional variational (4D-Var) data assimilation system. Finally, the relative sensitivity of a baroclinic instability wave development is obtained and shown using the MPAS-A adjoint model.

中文翻译：

---

MPAS-大气动态核心的切线线性和伴随模型的发展及其在伴随相对灵敏度研究中的应用

摘要 本研究开发并测试了 Python 驱动的非流体静力模型，用于跨尺度预测 - 大气 (MPAS-A) 动态模型，以及其切线线性和伴随模型。MPAS-A 的非线性、非流体静力动态核心被重组为具有 Python 驱动程序，以方便解析名称列表、操作矩阵、控制仿真时间流、读取模型输入和写入输出，而重型中介层和模型层保留在 Fortran 中以提高计算效率。在相同的Python驱动结构下，开发了MPAS-A模型动态核心的切线线性和伴随模型，并验证了正确性。Jablonowski 和 Williamson 的斜压波案例用于证明 MPAS-A 切线线性模型的近似精度和 MPAS-A 伴随模型对相对灵敏度研究的适用性。数值实验结果表明，切线线性模型可以很好地逼近四天预测期内所有模型变量的非线性模型扰动的时间演变。使用 MPAS-A 伴随模型，表明 24 小时地表压力预报的最敏感区域与天气有关。一个有趣的向西垂直倾斜也在一个 24 小时的表面压力预测的相对灵敏度结果中发现，该结果位于槽内的一个点以模拟初始条件。MPAS-A 伴随模型的这种功能对于理解动力学和变分数据同化非常重要。简明语言摘要 MPAS-A 是一种先进的全球数值天气预报模型，具有六边形网格，可以在某些感兴趣的目标区域压缩以获得更高的分辨率，并在其他目标区域平滑过渡到粗分辨率。在这项研究中，首先开发了 Python 驱动的 MPAS-A 模型，结合了灵活的 Python 驱动程序和 Fortran 的快速计算，使 MPAS-A 模型对用户和平台非常友好。然后开发了 MPAS-A 动力核心的切线线性和伴随模型，这两种模型都是各种灵敏度研究所必需的。它们也是未来基于 MPAS 的全球四维变分 (4D-Var) 数据同化系统不可或缺的组成部分。