Model development

Contributions to the development of COSMO and IFS

In the framework of the PASC-funded project external page KILOS (Kilometer-scale nonhydrostatic global weather forecasting with IFS-FVM) our group is currently involved in the further development of a new version of the Integrated Forecasting System (IFS) of the ECMWF. Instead of the spectral transform technique, this IFS version is based on finite volume spatial discretization, and the coding implementation of the dynamical core is based on GT4Py – a Python framework that includes a high-level embedded DSL to write stencil computations – which will enable access to heterogeneous supercomputing architectures, performance portability and scientific productivity. This project is co-lead by external page Christian Kühnlein at the ECMWF and supported by other colleagues at the ECMWF and CSCS. The main project scientists are Stefano Ubbiali, Nicolai Krieger and Lukas Papritz.

Earlier model developments of the group focused on adding atmospheric water cycle diagnostics to the limited-area COSMO model. COSMO-tag includes the tagging of pre-specified moisture sources and can be used to quantitatively analyze the contributions of different sources to, e.g., a heavy precipitation event. COSMO-iso is a model version equipped with stable water isotopes and can simulate d2H and d18O in vapour and precipitation. Both models were applied for studies in different geographical regions, and the basic implementations are documented in the publications by Winschall et al. 2014 (Atmos. Chem. Phys., 14, 6605-6619) for COSMO-tag and Pfahl et al. 2012 (Atmos. Chem. Phys., 12, 1629-1648) for COSMO-iso. Both these developments were mainly led by external page Stephan Pfahl.

The PhD projects conducted in this research are are outlined below:

Coupling parameterisations to the dynamical core

Enlarged view: Moist baroclinic wave simulation after 10 days of integration time. The top row shows the total column liquid water for CLOUDSC-dwarf (a), and the difference in total column liquid water between CLOUDSC-cy49r1 and CLOUDSC-dwarf (b). The panels (c) and (d) in the bottom row are, respectively, the same as (a) and (b), but for the total column ice water. The black contour lines indicate surface precipitation, 3 mm h-1 for the outer larger areas, and 30 mm h-1 for the inner smaller areas. The grey arrows denote surface winds stronger than 15 m s-1. — Moist baroclinic wave simulation after 10 days of integration time. The top row shows the total column liquid water for CLOUDSC-dwarf (a) and the difference in total column liquid water between CLOUDSC-cy49r1 and CLOUDSC-dwarf (b). The panels (c) and (d) in the bottom row are the same as (a) and (b) but for the total column ice water. The black contour lines indicate surface precipitation, 3 mm h^-1 for the outer larger areas and 30 mm h^-1 for the inner smaller areas. The grey arrows denote surface winds stronger than 15 m s^-1.

Gabriel Vollenweider

The European Centre for Medium-Range Weather Forecasts (ECMWF) has developed a new finite volume module (FVM) for the Integrated Forecasting System (IFS). The IFS-FVM was originally implemented in Fortran, but has also been ported to Python using the domain-specific library GT4Py to achieve high-performance on a variety of supercomputers. In ongoing efforts, the Python code is extended with the latest versions of the physical parameterisations from the IFS. The ultimate goal is to have a complete model in Python, including both the finite volume dynamical core IFS-FVM and the physical parameterisations from the IFS. This full model is termed the Portable Model for Multi-Scale Atmospheric Prediction (PMAP).

At high-resolutions between 1 km and 100 m, the main processes that need to be parameterised are atmospheric turbulence, cloud microphysics, and radiative transfer. The PMAP model already includes an atmospheric turbulence scheme and an older version of the IFS cloud microphysics scheme, called CLOUDSC-dwarf. In the first part of this project, the CLOUDSC-dwarf parameterisation is updated to the most recent version CLOUDSC-cy49r1, and the IFS radiation scheme ecRad is added to PMAP. A comparison of the different CLOUDSC versions is shown in the Figure above.

The second part of the project is concerned with the question of how to optimally couple the different parameterisations to the dynamical core. In the real atmosphere the different processes are tightly linked and calling them one after the other may neglect important feedbacks among different atmospheric scales. Calling the processes once in the forward order and once in the backward order may improve the numerical treatment of such interactions. To compare several possible coupling strategies, ensemble simulations of real case weather systems are run and evaluated against each other or against observational data.

In the final part of the project the optimal coupling strategy is selected to run a high-resolution PMAP simulation of a challenging weather phenomenon. The strengths of PMAP are analysed by comparing the PMAP forecast to forecasts from other numerical weather prediction systems.

Supervised by Heini Wernli and Stefano Ubbiali

High-resolution simulations with IFS-FVM

Enlarged view: Comparison of simulated and analytical solutions of idealized flow past topography. Shown are in shading simulated (a) vertical velocity and (b) perturbation of horizontal velocity along the vertical cross-sections. The contours show a linearised analytical solution on the same grid as the simulations. — Comparison of simulated and analytical solutions of idealized flow past topography. Shown are in shading simulated (a) vertical velocity and (b) perturbation of horizontal velocity along the vertical cross-sections. The contours show a linearised analytical solution on the same grid as the simulations.

Nicolai Krieger

The finite volume module (FVM) of the Integrated Forecasting System (IFS) is a new dynamical core that solves the fully compressible nonhydrostatic all scale governing equations based on a generalized height-based terrain-following vertical coordinate. The integration is done with non-oscillatory forward-in-time Eulerian advection and a centred two-time-level semi-implicit integration scheme. Through a finite-volume discretization and an implementation in Python with the domain specific language GT4Py, the IFS-FVM addresses challenges of heterogeneous and diverse high performance computing architectures and the need to apply numerical techniques minimizing the communication footprint of the model.

This project will contribute to the further development of the IFS-FVM. The IFS-FVM will be prepared in different configurations for high-resolution numerical experiments to test and validate the IFS-FVM. The main part of this project will then use IFS-FVM’s capabilities to simulate atmospheric flow phenomena from implicit large-eddy simulations to global convection-resolving simulations to gain novel insight into interesting research questions in atmospheric dynamics. Using large-eddy simulations, local wind phenomena such as the Laseyer wind in the mountains of north-eastern Switzerland will be investigated. The influence of different flow parameters such as environmental wind speed and wind direction on metrics of the flow in the target region will be investigated. Towards the end of the project, global convection-permitting simulations of extreme precipitation related to atmospheric rivers or intense warm conveyor belts will be performed and investigated.

Supervised by Heini Wernli, external page Christian Kühnlein (ECMWF), and Hanna Joos