Extending Forecast Reliability
Traditional physics-based weather forecasting and newer artificial intelligence models typically become less reliable beyond two weeks, a limitation that researchers aim to address through machine learning.
Microsoft Has a New Machine Learning Tool to Improve Weather Forecasts
New open-source software aims to make AI-enhanced weather forecasting more accessible and affordable for resource-constrained agencies.
Researchers report a cost-effective method to improve weather forecasts two to six weeks in advance, potentially helping farmers and emergency responders plan ahead. The findings were published in the journal Nature, and the underlying software is now open source.
Correcting Model Bias
Microsoft joins other major technology companies developing artificial intelligence tools for meteorology. Rather than building a standalone forecasting model, the new approach utilizes machine learning to correct systemic biases in existing models.
It offers a more accessible way for resource-strapped weather agencies to improve prediction accuracy using artificial intelligence.
While physics-based modeling requires supercomputers and generative AI depends on powerful graphics processing units, researchers note the new tool can operate using standard CPUs. Developers emphasize the potential for increased resource efficiency and extended reliability for extreme weather predictions.
Both traditional and AI models encounter challenges when predicting weather more than two weeks ahead. Models rely on approximations to fill data gaps from sparse stations or complex atmospheric phenomena, leading to systematic errors that compound over time.
The method, called Probabilistic Bias Correction, identifies systematic errors in a model and applies automated corrections. For example, the tool can adjust for climate change factors when historical records bias predictions toward colder temperatures.
Performance and Community Reception
In testing, the framework improved the accuracy of a leading physics-based model from the European Centre for Medium-Range Weather Forecasts across the vast majority of precipitation and temperature targets, while also doubling the accuracy of the centre's AI forecasting system.
Collaborators from various institutions describe the framework as a positive contribution to the meteorological community, noting that operational implementation remains a key challenge.
Turning conceptual ideas into an operational process is considered a challenging step.
Accessibility for Developing Regions
High infrastructure requirements have historically kept advanced weather forecasting out of reach for agencies in economically developing countries. The lightweight nature of the new models allows scientists with limited resources to configure and optimize them for local conditions.
Researchers are currently testing the framework with meteorological services in regions vulnerable to severe weather events and extreme seasonal variations.
