Microsoft Develops Machine Learning Tool to Improve Weather Forecasts
Microsoft has developed an open-source tool that improves the accuracy of forward-looking forecasts by correcting systematic biases in existing weather prediction models.
Microsoft researchers have announced the development of a new machine learning tool called Probabilistic Bias Correction, which makes weather forecasts between two and six weeks ahead more reliable.
The Two-Week Limit in Weather Forecasting
Traditional physics-based weather forecasts and new artificial intelligence models typically lose reliability in predictions extending beyond two weeks.
Company researchers state that they have found a cost-effective solution that will enable farmers and emergency response teams to make safer plans.
Correcting Biases in Existing Models
Instead of building an independent new weather model, Microsoft utilizes machine learning to correct biases in existing models.
This approach offers a more accessible way for resource-constrained weather agencies to improve their accuracy rates with the help of artificial intelligence.
An Alternative to High-Cost Systems
While physics-based modeling requires supercomputers, generative artificial intelligence needs powerful graphics processing units.
The tool exhibits a structure that is lightweight and efficient enough to run solely on CPUs without the need for expensive hardware.
Probabilistic Bias Correction Method
This method by Microsoft, called Probabilistic Bias Correction, uses machine learning to identify and correct systematic errors in a model.
Climate changes, such as greenhouse gas emissions warming the Earth, can cause models based on historical data to make colder predictions compared to reality.
Achieved Success Rates
Researchers found that the developed tool significantly increased the accuracy of the physics-based model of the European Centre for Medium-Range Weather Forecasts.
The system managed to increase prediction success rates in 98 percent of precipitation targets and 92 percent of temperature targets.
Opportunity for Developing Countries
Due to necessary infrastructure costs, technological advancements are often out of reach for agencies in economically developing countries.
The authors emphasize that these lightweight models allow scientists to make local-scale improvements without requiring massive resources.