Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Rain Field Reconstruction using Commercial Microwave Links and Diffusion Model Priors
Badr MOUFAD, Albina Ilina, Hai Victor Habi, Salem Lahlou, Yazid Janati, HAGIT MESSER, Eric Moulines
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| model the nonlinear power-law path-integrated operator instead of assuming b=1 (ablation) | MGPS with diffusion prior · Nonlinear observations, strong prior | RMSE: 0.96 -> 0.86 | Improved |
| model the nonlinear operator instead of assuming b=1 (ablation) | TDS with diffusion prior · Nonlinear observations, strong prior | RMSE: 0.97 -> 0.84 | Improved |
| use a strong (fully trained) diffusion prior instead of an under-trained one, with nonlinear operator (ablation) | MGPS · Nonlinear observations | RMSE: 0.88 -> 0.86 | Improved |
| use a strong prior instead of weak prior under the linear operator (ablation) | MGPS · Nonlinear observations, linear inference | RMSE: 0.94 -> 0.96 | No clear effect |
| diffusion posterior sampling instead of interpolation baselines (IDW, GMZ, OK) | CML rain reconstruction · OpenMRG real CML links | RMSE / PCC: OK 0.87 -> MGPS 0.86, TDS 0.84 (isotropic RMSE) | Improved |
| use diffusion methods in an idealized linear setting with dense short links (ablation) | diffusion posterior samplers vs interpolation · OpenMRG simulated CMLs, b=1 | RMSE: OK 0.83 vs MGPS 0.84 (few-long); IDW 0.75 vs MGPS 0.75 (many-short) | No clear effect |
| increase number of short CMLs (many-short vs few-long) (ablation) | IDW · OpenMRG simulated CMLs, b=1 | RMSE: 0.89 -> 0.75 | Improved |
| heteroscedastic noise (lower overall noise) instead of isotropic noise (ablation) | MGPS · OpenMRG real CML links | RMSE: 0.86 -> 0.83 | Improved |
| choose sampler TDS vs REDDIFF on GP benchmark | diffusion posterior samplers · GP 1D linear inverse problem | Sliced Wasserstein: 0.38 -> 0.07 | Improved |
| MGPS vs other samplers on cumulative rain bias | real CML reconstruction · OpenMRG real CML links (isotropic) | cumulative rain error: 42 vs -320 (DAPS), -356 (REDDIFF) | Improved |
Closest papers by meaning
- PrecipDiff: Leveraging Image Diffusion Models to Enhance Satellite-Based Precipitation Observations (AAAI 2025 · similarity 0.85)
- Coupled Data and Measurement Space Dynamics for Enhanced Diffusion Posterior Sampling (NeurIPS 2025 · similarity 0.84)
- Satellite Observations Guided Diffusion Model for Accurate Meteorological States at Arbitrary Resolution (CVPR 2025 · similarity 0.84)
- Precipitation Downscaling with Spatiotemporal Video Diffusion (NeurIPS 2024 · similarity 0.83)
- DiffCast: A Unified Framework via Residual Diffusion for Precipitation Nowcasting (CVPR 2024 · similarity 0.83)
- Analyzing and Guiding Zero-Shot Posterior Sampling in Diffusion Models (NeurIPS 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
