Microsoft Reveals Machine Learning Tool for Weather Forecasts
Microsoft has introduced a new machine learning software designed to tackle the reliability limits of extended weather forecasts.
Radwan AbdulhadiEditor and technology writer
Extending Forecast Reliability Beyond Two Weeks
Traditional physics-based weather forecasting systems and even newer AI models typically experience a drop in reliability when predicting weather beyond a two-week window. Microsoft researchers claim to have addressed this limitation by developing a new, cost-effective machine learning approach that enables more accurate weather forecasts two to six weeks in advance. This advancement aims to give farmers and emergency responders greater certainty when planning ahead.
The research findings regarding this system were published today in the journal Nature, and the underlying technology has been released as open-source software.
Correction Bias Instead of Standalone Models
With this announcement, Microsoft joins Google, Nvidia, and other companies developing AI tools for weather forecasting. Rather than constructing an entirely new standalone weather model, Microsoft’s research utilizes machine learning to correct systemic biases present in existing weather models. The researchers hope this provides a more accessible pathway for resource-constrained weather agencies to leverage artificial intelligence for improved prediction accuracy.
Lightweight Infrastructure Requirements
While traditional physics-based forecasting requires supercomputers and generative AI relies heavily on powerful GPUs, Microsoft Research New England senior principal researcher Lester Mackey stated that the new tool can operate exclusively on standard CPUs. Mackey noted optimism regarding both the enhanced resource efficiency of weather forecasting and the extended reliability of predictions, particularly when dealing with extreme weather events.
Understanding Probabilistic Bias Correction (PBC)
Both physics and AI models encounter complications when attempting forecasts two weeks or further into the future. Traditional systems simulate atmospheric physics, while AI models analyze historical data patterns. Both methodologies rely on approximations to address data gaps from sparsely monitored regions or complex atmospheric phenomena, creating systematic biases that compound over time.
Microsoft's solution is called Probabilistic Bias Correction (PBC). The software identifies systematic biases within a model and applies adjustments. For example, greenhouse gas emissions from fossil fuels continue to warm the planet, but weather models relying strictly on older historical records can suffer from a cold temperature bias.
Performance and Community Reception
In testing, researchers discovered that PBC enhanced the accuracy of a leading physics-based weather model from the European Centre for Medium-Range Weather Forecasts (ECMWF) across 98 percent of precipitation targets and 92 percent of temperature targets. Furthermore, the machine learning framework doubled the accuracy of the ECMWF's new AI forecasting system during the analysis. Prior to publication, the ECMWF recognized the research team's work on PBC through its AI Weather Quest competition.
Matthew Chantry, the ECMWF's strategic lead for machine learning, remarked that the paper serves as a strong example of community-oriented innovation, while also emphasizing that transitioning concepts into operational reality remains a challenging step.
Accessibility for Developing Regions
The massive computing infrastructure historically required for advanced weather forecasting has often placed these technological improvements out of reach for agencies in economically developing countries. However, the study's authors believe PBC offers a different approach due to its lightweight design and straightforward training process, enabling scientists with fewer resources to configure models for localized performance.
Genevieve Flaspohler, a co-author of the paper who worked on PBC as a Microsoft intern and now serves as executive director of Rhiza Research, is collaborating with the Kenya Meteorological Service Authority. Supported by funding, the organization is testing AI approaches to deliver accurate early warnings ahead of expected severe weather conditions.
Sources
- The VergeMicrosoft has a new machine learning tool to improve weather forecasts
Continue chronologically
Related entity coverage
Related stories
NVIDIA Opens Vera Rubin Architecture to Scale AI Factories
NVIDIA unveiled key rack and compute tray designs for the Vera Rubin architecture at the OCP Global Summit to help scale global AI factories.
Google's SynthID AI Detector Goes Public for Everyone
Google's SynthID Detector is now available globally for regular users, offering a simple way to check if images, videos, or audio files are AI-generated.
Googlebooks Face Tough Market Challenges Amid High Pricing
Google's new Googlebooks platform faces a difficult market launch, carrying a $999 starting price and a strict 16GB RAM requirement during rising component costs.
Microsoft Unveils Third-Generation Maia 300 AI Chip
Microsoft announces its third-generation Maia 300 in-house AI chip, designed to improve cloud efficiency and reduce reliance on Nvidia GPUs.
