Awesome Diffusion Models for Weather Forecasting

This repository serves as a curated collection of research papers focusing on the application of diffusion models to weather forecasting. It targets researchers, data scientists, and engineers in atmospheric science and machine learning seeking to leverage cutting-edge generative AI for meteorological predictions. The primary benefit is providing a centralized, up-to-date resource to track advancements and identify potential solutions for a wide array of weather prediction challenges.

How It Works

The core approach documented across the listed papers involves applying diffusion models, a class of generative AI, to various weather forecasting tasks. These models learn to iteratively denoise data, enabling them to generate realistic weather patterns, downscale resolutions, emulate complex climate models, and provide probabilistic forecasts with uncertainty quantification. This generative capability offers novel solutions for tasks like precipitation nowcasting, extreme event prediction, and data assimilation, often capturing complex spatio-temporal dependencies.

Quick Start & Requirements

This repository is a curated list of research papers and associated code. To "get started," users should browse the listed papers and access their linked code repositories for specific installation and execution instructions. Prerequisites and dependencies (e.g., Python versions, deep learning frameworks, hardware like GPUs) are unique to each individual project and must be consulted from the respective linked repositories. Links to papers and code are provided for each entry.

Highlighted Details

• Comprehensive coverage of diffusion model applications in atmospheric science, spanning precipitation nowcasting, global forecasting, downscaling, data assimilation, emulation, and extreme weather prediction.
• Inclusion of direct links to research papers and, frequently, to associated code repositories, facilitating reproducibility and further investigation.
• Chronological ordering of papers and a commitment to weekly updates (via Arxiv checks) ensure a relatively current overview of the field.
• Demonstrates the versatility of diffusion models in addressing diverse and challenging weather forecasting problems.

Maintenance & Community

The repository is actively maintained by a single individual who performs weekly checks of Arxiv for new diffusion-related atmospheric science papers. No specific community channels (e.g., Discord, Slack) or contributor information are provided.

Licensing & Compatibility

The repository itself does not specify a license. Users must refer to the individual research papers and their linked code repositories for licensing terms and compatibility, particularly for commercial use.

Limitations & Caveats

As a curated list, this repository does not offer a unified software framework or API. The maintenance is manual, potentially leading to missed publications. Accessing and running the associated code requires navigating the distinct setup, dependencies, and licensing of each individual project.