Get Started
Home
Topics
Search
Library
Research questionHow can we improve weather forecasts two to six weeks ahead despite rapidly degrading skill?Forecast accuracy falls sharply beyond two weeks as forecast errors compound, systematic biases persist, and atmospheric chaos limits predictability. This makes forecasts less reliable for agricultural, energy, water, wildfire, and disaster-preparedness decisions.
AI
Diffusion Models
Machine Learning
Research Paper
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.Advancing Subseasonal Forecasting with Machine LearningThe source examines probabilistic bias correction learned from historical forecasts and applied to ECMWF dynamical and AI forecasting systems. Evidence covers global pressure, temperature, and precipitation forecasts and reports results from ECMWF’s 2025 real-time forecasting competition.research paper · Sep 4, 2026SimCast-S2S: A Computationally Efficient Diffusion Model for Subseasonal Precipitation ForecastingThe evidence concerns SimCast-S2S, a latent-diffusion system evaluated on reanalysis data. It uses variational-autoencoder latent representations and transfer learning from climate simulations with low-rank adaptation; the reported results use a subset of atmospheric variables without post-processing, bias correction, or calibration, and compare against deep-learning baselines and the ECMWF-S2S baseline.research paper · Sep 3, 2026
Related questions
How can multimodal models improve tropical-cyclone track and intensity forecasts across weather foundation models without retraining?How can AI weather models produce high-resolution global forecasts without relying solely on analysis data?How can cloud-cover forecasts extend beyond nowcasting despite imperfect NWP initialization while preserving satellite spatial detail?How can probabilistic weather downscaling recover site-specific conditions from coarse grids despite unresolved terrain and land-surface effects?