Higher resolution is expected to improve representability of simulations especially in location and timing of localized heavy precipitation. For such high-resolution simulation, efficient use of computational resources, such as CPU, memory, storage, network, and so on, are a crucial issue, and collaboration with computer scientists is required.

Understanding a sensitivity and validity of schemes in climate and weather simulation is also important for high-resolution simulations. In the climate/meteorological model, most of processes, especially physical processes, are based on empirical rules and hypotheses. This leads to large uncertainties in simulation result. To evaluate the reliability of the simulations, direct comparison between various schemes is necessary through conducting the sensitivity of schemes. So far, through many inter-model comparison studies we have obtained a lot of fruitful knowledge. However, it is difficult to clarify what is an origin that bears the difference of results, because there are many differences in the components and schemes between the models.

For future high-resolution simulations having such issues, we have developed a basic library and environment for meteorological numerical simulation models, named SCALE (Scalable Computing for Advanced Library and Environment; Nishizawa et al 2015, Sato et al 2015) published on our web site (http://scale.aics.riken.jp/) as an open source software.

Collaborating with computer scientists in AICS, we have made an effort for better scalability and higher efficiency of computation in recent and future HPC (High Performance Computing) systems. The higher-resolution simulation still requires appropriate choice or modification of the schemes and parameterizations, and we have tried to evaluate schemes in high-resolution simulations.


Slides to this talk