High‐Precision and Fast Prediction of Regional Wind Fields in Near Space Using Neural‐Network Approximation of Operators

Abstract

Abstract Fine modeling and fast prediction of regional wind field in the middle and upper atmosphere has always been a difficult problem. We designed a neural operator method to solve this problem. We combine the idea of data assimilation with deep learning method to design a regional wind field operator suitable for near space. The annual Root mean square error of the zonal wind and meridional wind of the operator model at the height of 30 km are 0.903 and 0.881, respectively, which is three times that of ConvLSTM. Moreover, we validate the sparse spatio‐temporal modeling method of regional wind field operator at 20/30/40/50 km altitude. The result shows that the model is mesh‐free, and can get high‐precision modeling of different spatio‐temporal resolutions, multiple regions and arbitrary positions at one time, which lays an foundation for fine regional modeling and rapid utilization of near space.

Keywords

• near space
• regional wind fields
• spatio‐temporal modeling
• neural operator