AOD Reconstruction with Uncertainty via Diffusion Models
Published:Dec 31, 2025 13:16
•1 min read
•ArXiv
Analysis
This paper addresses the challenge of reconstructing Aerosol Optical Depth (AOD) fields, crucial for atmospheric monitoring, by proposing a novel probabilistic framework called AODDiff. The key innovation lies in using diffusion-based Bayesian inference to handle incomplete data and provide uncertainty quantification, which are limitations of existing models. The framework's ability to adapt to various reconstruction tasks without retraining and its focus on spatial spectral fidelity are significant contributions.
Key Takeaways
- •Proposes AODDiff, a probabilistic reconstruction framework for AOD fields.
- •Utilizes diffusion-based Bayesian inference to handle incomplete data and quantify uncertainty.
- •Employs a corruption-aware training strategy and decoupled annealing posterior sampling.
- •Demonstrates efficacy in downscaling and inpainting tasks, maintaining high spatial spectral fidelity.
- •Offers uncertainty quantification via multiple sampling, providing confidence metrics.
Reference
“AODDiff inherently enables uncertainty quantification via multiple sampling, offering critical confidence metrics for downstream applications.”