Super‐Resolution Reconstruction of Near‐Space Global‐Scale Temperature Fields (50–80 km) Utilizing an Extendable Multi‐Input Operator Neural Network Based on Reanalysis Data and Satellite Observations

Abstract The reanalysis data set between 50 and 80 km altitude suffers from its low vertical resolution and high uncertainty errors. This paper introduces an extendable multi‐input operator neural network, designed to explore global‐scale temperature field to achieve super‐resolution reconstruction in this region. This architecture demonstrates the ability to flexibly address high‐dimensional problems of multi‐source data. Utilizing reanalysis data and observations, the super‐resolution operator elevates the vertical stratification of the temperature field from 13 to 31 layers, achieving a vertical resolution of 1 km, while correcting errors. The reconstructed data set demonstrated a reduction in root mean square error and mean absolute error metrics by 19.3% and 25.1%, respectively. These improvements are particularly pronounced between 55 and 70 km altitude. Notably, the super‐resolution operator model exhibits mediocre performance at heights above 70 km. Our research offers novel insights into generating high‐fidelity near‐space atmospheric data.