Taiwan Weather Agency Adopts AI Models to Sharply Improve Typhoon Forecasting

Taiwan’s Central Weather Administration previously relied mainly on conventional physics-based models, using supercomputers to calculate atmospheric changes. A single typhoon simulation took several hours. By introducing AI models powered by GPUs and data-driven technology, the agency can now generate multiple storm-track forecasts more quickly, buying critical time for typhoon warnings, evacuations and disaster-response decisions. Reports did not disclose the cost of the deployment.

The agency has deployed 15 global AI weather models, cutting the time needed to simulate typhoon movements from several hours to about 10 minutes. Forecast accuracy has improved by an amount equivalent to four years of progress in conventional models. The system was used to forecast Typhoon Gaemi in July 2024, with preliminary results demonstrating the AI models’ practical value in computational efficiency and storm-track analysis.