BEIJING — China has built a weather forecasting system that predicts conditions up to 10 days ahead at a resolution of 5 kilometers, using 100,000 domestically made artificial intelligence chips. The resolution is finer than that of the leading U.S. global forecast model, showing that China is accelerating technological self-reliance in weather forecasting as well, on the back of homegrown AI infrastructure.
The China Meteorological Administration and Chinese computing firm Sugon said recently that they had produced a 10-day global weather forecast at 5-kilometer resolution using Sugon 8000, a large-scale computing cluster built with domestic AI chips, according to the South China Morning Post on the 17th.
Most major global weather forecasting systems currently operate at resolutions of roughly 5 to 10 kilometers. The Global Forecast System (GFS), run by the U.S. National Weather Service, has a maximum resolution of about 13 kilometers. Chinese researchers have already succeeded in tests at 3-kilometer resolution, and they aim to achieve forecast accuracy of 1 kilometer nationwide and 100 meters in key regions by 2030.
Raising forecast resolution requires enormous computing power, because dividing the atmosphere into a finer grid sharply increases the volume of calculations. Halving the distance between grid points, for example, can increase the required computation by up to eightfold.
For this, China used Sugon 8000, unveiled in July. The system is a large-scale computing cluster linking 100,000 high-performance computing units, built with Chinese-made chips, storage devices and networking technology. Its defining feature is a design that allows high-precision scientific computation, a traditional strength of supercomputers, and the large-scale data processing capability of AI systems to run simultaneously on a single system.
In the test, Sugon 8000 calculated the 10-day global forecast in about an hour. That is not faster in absolute terms — the U.S. GFS has processed a 10-day forecast in about 45 minutes in a test environment — but Sugon said the system met the timing requirements for routine forecasting operations.
China expects the system to strengthen its ability to respond to extreme weather such as typhoons and torrential rain. Beyond weather forecasting, Sugon 8000 is being used for protein research, molecular dynamics and turbulence simulation, and it currently supports about 200 applications and more than 500 AI models across more than 30 research and industrial fields.
