Spatiotemporal Prediction of Global Atmospheric Duct Height Across Latitudes Based on ERA5 Data

Accurate prediction of global atmospheric duct heights is important for a variety of purposes, including communications, navigation, meteorological forecasting and the military. However, the global atmospheric duct height data is wide-ranging and costly for overall spatiotemporal prediction. In addition, the traditional convolutional neural network (CNN) and recurrent neural network (RNN) have some limitations and cannot extract spatial and temporal features efficiently at the same time. In this paper, an efficient spatiotemporal prediction network for global atmospheric duct heights combining 3D convolutional neural network and Transformer is proposed based on ERA5 reanalysis data. Considering the spatial heterogeneity of atmospheric duct distribution latitude, the global data are divided into three latitude regions of low, middle and high latitude for spatiotemporal prediction respectively, and the prediction results are fused to finally obtain the spatiotemporal prediction results of the global atmospheric duct height. The experimental results show that compared with the overall prediction method, the deep spatiotemporal prediction model with cross-latitude fusion proposed in this study has a significant improvement in speed and accuracy, and outperforms other existing networks.