In March 2023, the weather forecast monitoring company ForecastWatch gave the Microsoft Start weather development team the title "World's most accurate global forecast provider." It seems the Microsoft Start team is not resting on its laurels. This week, Microsoft announced that the team had come up with a more accurate way to predict the weather as far as 30 days in advance.
In a post on the Bing blog, Microsoft says its Start team published a paper on a new medium-range forecast model on the arXiv site at Cornell University. It shows how this new model is compared to the current one used by the European Centre for Medium-range Weather Forecasts (ECMWF). In summary, the new Microsoft system combines the use of five different AI models and three deep learning architectures to come up with their forecasts.
Microsoft says that these AI models can take decades of weather data so they can know familiar patterns in the present day and then predict what they will do in the future. It added:
They operate in much the same way as an NWP model, though: given the current state of the atmosphere on a 3-D globe (latitude, longitude, and height), predict the state of the atmosphere for some future time, say one hour later. They then feed this prediction back into the model to predict two hours later, and so on.
Microsoft's AI models can run predictions faster with the use of a GPU, and as a result, they can be run more frequently. In theory, that should result in better forecasts.
So far, Microsoft says its methods for measuring temperature errors are more accurate than the ones used by the ECMWF by 17 percent for one-week weather forecasts and by 4 percent for four-week forecasts. The company will be adding this model for 30-day weather forecasts into its Microsoft Start modeling.
6 Comments - Add comment