Presentation Information
[O13-P03]Machine Learning-Based Local Upper-Wind Prediction Using Reanalysis Atmospheric Data
*Ritsuka Shoyama1, Mahito Sugiyama2 (1. Shibuya Kyoiku Gakuen Makuhari Senior High School, 2. National Institute of Informatics)
Keywords:
Upper wind,Prediction,Machine learning,Reanalysis Atmospheric Data,Rocket,Balloon
Accurate upper-atmosphere wind data is essential for trajectory calculation and safety evaluation in rocket and balloon launches. However, current operations rely on radiosonde observations obtained on the launch day, making it difficult to make advance decisions about whether to proceed with the launch or postpone it. Although prediction models can be used to forecast weather, their coarse spatial resolution and high computational cost limit their applicability for local and real-time predictions. To address this issue, we develop a machine learning–based method that directly predicts local upper-atmosphere wind conditions in a cost-effective and timely manner for launch planning. In our previous study, we trained the model using five years of upper-air observation data from the University of Wyoming, limited to pressure levels down to 150 hPa. In this study, we introduce reanalysis data covering pressure levels down to 1 hPa without missing values, enabling prediction across a wider altitude range. We apply the same regression-based approach and evaluate the predictive performance using the root mean square error (RMSE) of wind vectors at each pressure level. The results show that prediction accuracy near the surface improved compared to the previous model. Up to 150 hPa, the accuracy remained comparable despite using only one year of training data. At higher altitudes, the error varied depending on the pressure level. These results demonstrate that machine learning combined with reanalysis data provides a practical and scalable approach for local upper-atmosphere wind prediction, contributing to more reliable pre-launch decision-making.