Weather Forecasting
Overview
Weather forecasting estimates how atmospheric conditions will evolve to support decisions ranging from daily planning to severe-weather warnings. Numerical weather prediction solves physical equations, while machine-learning approaches use transformers, graph networks, state-space models, and generative methods trained on reanalysis, satellite, radar, or station observations; data assimilation combines observations with model states. Current work also develops probabilistic ensembles, short-range radar nowcasts, regional downscaling, and forecasts for events such as tropical cyclones and heavy precipitation. A central challenge is producing reliable long-range and extreme-event forecasts while preserving physical consistency, representing local detail, and validating performance against real observations rather than reanalysis alone.
Latest papers
189 papers16 papers in the last four weeks, down 56% on the four weeks before. 0.1% of all new papers.
PDE foundation models are skillful AI weather emulators for the Martian atmosphere
Symmetric Composition of Anisotropic Operators for Global Subseasonal-to-Seasonal Climate Forecasting
Long-Range Distillation: Distilling 10,000 Years of Simulated Climate into Long Timestep AI Weather Models
student'' model is trained to forecast directly at long-range using a huge synthetic dataset generated by an autoregressive teacher'' model. Using the Deep Learning Earth System Model (DLESyM) as teacher, we generate over 10,000 years of simulated climate to train student models across a range of timescales. In perfect-model experiments, the distilled models outperform climatology and approach their teacher's skill while replacing hundreds of autoregressive steps with a single timestep. In the real world, both a diffusion model student fine-tuned on ERA5 and a linear inverse model student with mixed DLESyM-ERA5 training achieve S2S forecast skill competitive with the ECMWF IFS ensemble, with synthetic pre-training providing improvements beyond ERA5-only training. This demonstrates the benefit of synthetic pre-training for both deep learning and classical data-driven methods, and represents the first demonstration that real-world skill can be improved by training on AI-generated synthetic data in volumes far exceeding the observational record.