Microsoft Develops Machine Learning Tool to Enhance Long-Range Weather Forecasting
Microsoft has developed a new machine learning tool called Probabilistic Bias Correction (PBC) that improves the accuracy of long-range weather forecasts, up to six weeks in advance. The open-source software addresses systematic biases in existing physics-based and AI-driven models by identifying and correcting errors using historical data. This approach allows for more reliable predictions without requiring high-end computing resources like supercomputers or powerful GPUs. The tool was tested on the European Centre for Medium-Range Weather Forecasts (ECMWF) model, improving accuracy for 98% of precipitation targets and 92% of temperature forecasts. It also doubled the performance of ECMWF’s AI forecasting system in tests. The research collaboration involved Microsoft, MIT, and other institutions. Microsoft's solution is more resource-efficient than traditional models or generative AI approaches, making it accessible for weather agencies with limited resources.
Topics: machine_learning open_source_software weather_forecasting