-
- A43R-2372: Machine Learning Downscaling of Global AI Forecasts for Near-Surface Wind Prediction
-
Board 2372‚ Hall EFG (Poster Hall)NOLA CC
Author(s):Generic 'disconnected' Message
Mingshi Yang, University of Illinois at Urbana-Champaign (First Author, Presenting Author)
Wind energy operators need frequent and accurate forecasts of wind speed at the height of wind turbines (around 100 m above ground). These forecasts are important for scheduling power production, balancing the electric grid, and planning future energy needs. However, most global weather forecast models, including advanced AI-based models, only predict wind at a few fixed heights, such as near ground level (10 m) or at pressure levels high in the atmosphere. They also often produce forecasts only every 6 or 12 hours, which makes it difficult to use their data directly for wind energy applications.In this study, we developed a machine learning method that uses information from many atmospheric levels to estimate wind speed at both 10 m and 100 m. We tested the approach using forecasts of NeuralGCM, a global AI weather forecast system, and compared the results with numerical and data-driven weather models over the central U.S. wind belt in 2020. The downscaled forecasts matched or outperformed the models we evaluated and provided reliable estimates at 100 m, the height where wind turbines typically operate. This approach can also be applied to other forecast systems and may help improve wind energy planning and operations.
Scientific DisciplineSuggested ItinerariesNeighborhoodType
Enter Note
Go to previous page in this tab
Session
