AI Weather Model SkyCast Achieves 30-Day Ultra-Long-Range Forecasting: 40% Accuracy Improvement Over Traditional Models

AI Weather Model SkyCast Achieves 30-Day Ultra-Long-Range Forecasting

On March 15, 2029, weather AI company DeepWeather released its SkyCast large model, extending the effective weather forecast window from 10 days with traditional numerical models to 30 days. In comparison testing with the European Centre for Medium-Range Weather Forecasts (ECMWF), SkyCast achieved 40% higher temperature accuracy and 28% higher precipitation accuracy at the 15-day forecast horizon.

SkyCast uses a global-regional dual-layer architecture: the global layer processes atmospheric circulation patterns at 0.25-degree resolution, while the regional layer performs local downscaling at 1-kilometer resolution. The model was trained on 15 years of historical data from 12,000 weather stations, 8 meteorological satellites, and 200 ocean buoys, with 80 billion parameters.

DeepWeather CEO Priya Sharma said: "30-day forecasting is enormously valuable for agricultural planting decisions, aviation fuel planning, and disaster prevention. One of our early customers, a large agricultural cooperative, saved approximately $1.2 million in water resource costs by adjusting irrigation plans three weeks ahead of a drought."

MIT atmospheric science professor David Romps cautioned that 30-day forecasting remains near the "predictability boundary": "The inherent properties of chaotic atmospheric systems set a theoretical limit for daily forecasts beyond two weeks. SkyCast's breakthrough is noteworthy, but users need to understand that 30-day forecasts provide probabilistic trends rather than deterministic predictions."