Skillful and realistic precipitation forecasting with reinforced diffusion
Abstract
Accurate precipitation forecasting remains elusive as physics-based models struggle to resolve mesoscale convection and machine learning often yields physically inconsistent, blurred results that consistently underestimate extreme events. We present FitzRoy, the first system that simultaneously achieves state-of-the-art predictive skill and physical realism over the continental United States. Through a reinforced diffusion framework, where a conditional discrete diffusion model is refined via reinforcement learning, FitzRoy effectively resolves the long-standing trade-off between pointwise accuracy and structural fidelity. Across lead times up to 24 hours, FitzRoy consistently matches or exceeds state-of-the-art models while preserving structural realism and capturing extreme events. Its low-latency inference further enables deployment in real-time operational settings. In case studies such as the 2025 Texas Hill Country floods and the 2026 Missouri floods, FitzRoy signals record-breaking accumulations that existing models largely missed. By bridging the gap between realism and skill, FitzRoy establishes a new frontier for high-fidelity precipitation forecasting.
Related
- Huayu-Global: Skillful real-time quasi-global precipitation retrieval
- Skillful Data-Driven Subseasonal Soil Moisture Forecasting: Prospects and Limits for Flash Drought Prediction
- Shallow-to-deep velocity model building via diffusion models-Part II: Realistic scenarios
- Diffusion-Based Refinement for Kilometer-Scale Probabilistic Precipitation Nowcasting
- PCSDiff: Diffusion-Based Bias Correction and Super Resolution Toward Practical Operational Medium-Term Precipitation Forecast
BibTeX
@misc{sener2026skillful,
title = {{Skillful and realistic precipitation forecasting with reinforced diffusion}},
author = {Ozan Sener and Anand Avati and David Hafner and Jierui Lin and Laura Manduchi and Xiaoying Pang and Langwen Huang and Torsten Hoefler and Russ Schumacher and Vladlen Koltun},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.38775},
eprint = {2610.38775},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.38775},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Skillful and realistic precipitation forecasting with reinforced diffusion
AU - Ozan Sener
AU - Anand Avati
AU - David Hafner
AU - Jierui Lin
AU - Laura Manduchi
AU - Xiaoying Pang
AU - Langwen Huang
AU - Torsten Hoefler
AU - Russ Schumacher
AU - Vladlen Koltun
PY - 2026
DA - 2026/10/28
PB - arXiv
DO - 10.48550/arxiv.2610.38775
UR - https://arxiv.org/abs/2610.38775
AB - Accurate precipitation forecasting remains elusive as physics-based models struggle to resolve mesoscale convection and machine learning often yields physically inconsistent, blurred results that consistently underestimate extreme events. We present FitzRoy, the first system that simultaneously achieves state-of-the-art predictive skill and physical realism over the continental United States. Through a reinforced diffusion framework, where a conditional discrete diffusion model is refined via reinforcement learning, FitzRoy effectively resolves the long-standing trade-off between pointwise accuracy and structural fidelity. Across lead times up to 24 hours, FitzRoy consistently matches or exceeds state-of-the-art models while preserving structural realism and capturing extreme events. Its low-latency inference further enables deployment in real-time operational settings. In case studies such as the 2025 Texas Hill Country floods and the 2026 Missouri floods, FitzRoy signals record-breaking accumulations that existing models largely missed. By bridging the gap between realism and skill, FitzRoy establishes a new frontier for high-fidelity precipitation forecasting.
ER -