We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A Python-based approach to the physics-dynamics coupling in atmospheric models.
- Authors
Ubbiali, Stefano; Schär, Christoph; Schlemmer, Linda; Sawyer, William; Schulthess, Thomas C.
- Abstract
Atmospheric models are complex systems where a dynamical core solves the fluid-dynamicsequations and a bundle of physical parameterizations express the bulk effect of subgrid-scalephenomena upon the large-scale dynamics. The procedure which molds all the dynamical andphysical components into a coherent and comprehensive model is referred to as thephysics-dynamics coupling. Whereas parameterizations have been largely studied in isolation,the physics-dynamics coupling has historically received less attention. To a certain extent,this deficiency may be ascribed to the lack of flexibility, interoperability, and usability oftraditional frameworks. We propose a Python framework - code named tasmania - to ease the composition,configuration, simulation and monitoring of Earth system models. The framework features acomponent-based architecture, with each component being a Python class representing adynamical or physical process. As a result, the user is given fine-grained control on theexecution flow. Physical components must conform to sympl’s (System for Modelling Planets)primitives application programming interface (API). To facilitate the development ofdynamical kernels, tasmania provides an abstract base class (ABC) with intended supportfor multi-stage time-integrators (e.g., Runge-Kutta schemes) and partial operator splittingtechniques, which integrate slow and fast processes with large and multiple small time steps,respectively. To this end, a distinction between slow physics (calculated over the large timestep, outside of the dynamical core), intermediate physics (evaluated over the large time stepat every stage) and fast physics (computed over the shorter time step at each sub-step) ismade. Four coupling mechanisms (concurrent coupling, parallel splitting, sequential-splitting,and symmetrized sequential-splitting) are currently implemented; hybrid approaches arepossible. A simplified hydrostatic model in isentropic coordinates is used as proof-of-concept.Finite difference operators arising from the numerical discretization of the model areimplemented via GridTools4Py - a domain specific language (DSL) for stencil-basedcodes which offers a high-level entry point to the high-performance GridTools library.Early results, highlighting the variety of performance across the diverse coupling methods,are showcased.
- Subjects
ATMOSPHERIC models; DIFFERENCE operators; PYTHON programming language; PARAMETERIZATION; KERNEL functions
- Publication
Geophysical Research Abstracts, 2019, Vol 21, p1
- ISSN
1029-7006
- Publication type
Article