Refined Reconstruction of Multimodal Mesospheric Temperature Fields Based on Diffusion Model

Abstract Mesosphere is the core region of atmospheric wave activities, where fine temperature data are essential for analyzing its dynamic processes. We use an interpretable Data Diffusion Fusion Model to fuse sparse Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) observations with coarse‐resolution Modern‐Era Retrospective Analysis for Research and Applications V2. By combining spatiotemporal properties, F10.7 and Kp, the mesospheric temperature is reconstructed with high‐precision, increasing resolution by 5 times. Results show that Denoising Diffusion Fusion Model outputs incorporates more information from SABER data, improving the Peak Signal‐to‐Noise Ratio by 5.02 dB. Our reconstruction model demonstrates excellent global stability ( R 2 > 0.94). In the January 2019 Sudden Stratospheric Warming event, the model not only accurately reproduces the small‐scale features, but also significantly reduces the deviation from 21.7% to 14.9% with SABER data. This outcome fully proves its superior performance in handling extreme events. Importantly, even in the absence of observations, our model can still deliver high‐precision data set to support scientific investigations.