WEAVE: Weather-field Exchange for Atmospheric Variable Estimation

Temperature, water-vapour mixing ratio, and vertical velocity all describe the same evolving atmosphere, yet are commonly denoised independently. In this study, we introduce WEAVE (Weather-field Exchange for Atmospheric Variable Estimation), a residual DnCNN that combines field-specific pathways with shared multivariate context to reconstruct all three fields simultaneously. On processed ARM Southern Great Plains observations under matched 40\% Gaussian corruption, WEAVE outperforms a training-calibrated anisotropic Gaussian smoother across every displayed field and metric. Compared with three independent DnCNN baselines, the simultaneous model improves mixing-ratio RMSE, PSNR, and SSIM across all five matched initialisations, while temperature improves in three of five initialisations for RMSE and PSNR and in all five for SSIM. WEAVE also uses 11.23\% fewer parameters, achieves stronger spectral fidelity across all three fields, and retains information relevant to a vertical-motion regime diagnostic. Together, these results show that shared atmospheric structure can improve denoising with lower model complexity, providing a promising foundation for future multivariate reconstructions of atmospheric observa$tions.