Taiwan's Weather Forecast Gets a Supercharge: AI-Powered Supercomputer Coming to Hsinchu

Boosting Accuracy and Speed: CWA to Revolutionize Weather Prediction with AI

The Executive Yuan has given the green light to the Central Weather Administration (CWA) to significantly enhance weather forecasting capabilities in Taiwan. The plan involves upgrading a weather observation station in Hsinchu County into the Hsinchu Meteorology Science Park, which will house a state-of-the-art supercomputer server room dedicated to artificial intelligence (AI) applications.

This ambitious project is poised to revolutionize weather prediction. According to CWA Administrator Lu Kuo-cheng (呂國臣), the integration of AI will allow for more precise weather forecasts and the development of innovative weather-related applications. Currently, the CWA lacks the necessary computing power and physical space for a supercomputer. The planned server room will be built to withstand significant weight, capable of supporting 2 tonnes per square meter.

The choice of Hsinchu for the supercomputer's location is strategic, as the county boasts high-quality, high-speed Internet connectivity. Lu Kuo-cheng emphasized that data, computing power, and skilled personnel are the three essential elements for leveraging AI in weather forecasting.

“We have weather data accumulated over the past 100 years, but we need computing power to train large-scale weather forecast models and integrate them with existing models. We are working with the National Science and Technology Council and Nvidia Corp to learn new ways to develop high-resolution forecast models using AI,” he said.

The "Construction Project of High-Speed Computer for Weather Forecast" is scheduled for completion in 2027. The impact of AI is expected to be substantial, potentially increasing the accuracy of global weather forecast models by 6 percent and improving the accuracy of typhoon path projections within 120 hours by 12 percent, according to Lu Kuo-cheng.

Chang Bau-liang (張保亮), deputy director of the CWA Remote Sensing Division, highlighted the plan to progressively increase the percentage of data processed through graphic processing units, from 30 percent to 40 percent and eventually 60 percent. Furthermore, training time for atmospheric models is anticipated to be accelerated by a factor of 1,000 to 1,500, resulting in weather forecast results available within minutes.

In addition to the supercomputer project, the CWA plans to introduce wind force forecasts for coastal areas this summer and enhance monitoring of high-intensity short-duration rainfall by deploying weather radar systems in Yunlin and Yilan counties.