Microsoft Open-Sources Weather Tool That Improves 2-6 Week Forecasts by Up to 98%
Updated
Updated · The Verge · Oct 14
Microsoft Open-Sources Weather Tool That Improves 2-6 Week Forecasts by Up to 98%
3 articles · Updated · The Verge · Oct 14
Summary
Nature published Microsoft’s Probabilistic Bias Correction study, and the company released the software as open source to sharpen forecasts two to six weeks ahead.
PBC uses machine learning to correct systematic bias in existing weather models rather than replace them, letting agencies run the tool on CPUs instead of supercomputers or high-end GPUs.
Tests on ECMWF systems showed accuracy gains for 98% of precipitation targets and 92% of temperature targets, while doubling the accuracy of the center’s newer AI forecasting system.
Microsoft and its research partners say the lightweight approach could help lower-cost weather agencies, especially in developing countries, improve planning for farming, disasters and extreme rainy seasons.
Microsoft says its open-source PBC doubled accuracy for one AI weather system—but will those gains hold in real-world forecasts worldwide?
If AI can correct hidden biases in top weather models, could it improve warnings for heatwaves and floods faster than building new forecasting systems?