The United Kingdom Met Office’s Unified Model (UM) has been in constant use around the world for over 25 years, serving – as its name suggests – as a unified hub for immediate weather forecasts, global climate modeling, and everything in-between. But now, according to a paper published in the Asia-Pacific Journal of Atmospheric Sciences, the Korea Institute of Atmospheric Prediction Systems (KIAPS) is nearing deployment of a challenger to the UM – a model it’s calling the Korean Integrated Model – or “KIM” – system.
The Korea Meteorological Administration (KMA), like many major weather organizations, currently runs a modified version of the UM on its supercomputers. But in South Korea’s case, plans were quickly made to supplant the British model. The KMA implemented the UM in 2010, and within just one year, KIAPS had begun a nine-year project with the intent to develop a global atmospheric system model that could match or exceed the accuracy of the UM.
Development progress
According to the researchers, KIM is innovative in two major ways. First, it is the first global numerical weather prediction system configured over a cubed-sphere grid, which reduces the appearance of errors around the poles. Second, it employs a uniquely accurate and efficient spectral-element non-hydrostatic dynamical core.
The spectral method, while popular due to its high accuracy, does become less efficient with higher resolution in an HPC environment. As a result, KIAPS decided to employ a hybrid spectral-element approach, which uses a spectral approach for partial differential equations but defers to an element-based approach for the remainder of the model, increasing efficiency.
In recent tests, the authors write, KIM has achieved comparable results to the UM with no discernible deficiencies. In particular, the authors claim comparative advantages in prediction of precipitation over the Korean peninsula. Still, they acknowledge that the model currently has a few disadvantages, including high computational burdens in dynamics and poor forecasting in tropical regions.
Plans for deployment
The model’s developers currently plan to incorporate a regional model and a package for seasonal forecasts and climate studies. There is potential down the line to couple it to an ocean model to enable more robust climate applications, but the researchers don’t have immediate plans to do so.
KIM is scheduled to replace the KMA implementation of the UM in 2020, following a final upgrade to KIM in late 2019. If the KMA implements KIM as planned, KIM will likely run on the KMA’s current supercomputing infrastructure, comprised of three XC40 supercomputers with a combined 6.2 petaflops of peak performance and Aries interconnect. The Cray machines replaced KMA’s prior Cray XE6 system in 2015.
About the paper
The paper referenced in this article, “The Korean Integrated Model (KIM) System for Global Weather Forecasting,” was written by Song-You Hong, Young Cheol Kwon, Tae-Hun Kim, Jung-Eun Esther Kim, Suk-Jin Choi, In-Hyuk Kwon, Junghan Kim, Eun-Hee Lee, Rae-Seol Park, and Dong-Il Kim. It can be found in the 54th volume of the Asia-Pacific Journal of Atmospheric Sciences.