GoogleのGenCast AIモデルは天気予報業界に革命を起こすことを目指している

The Ardeche was flooded by deluges last October. This weekend, from December 7th to December 8th, violent winds ravaged northern France .

Today, we must simulate the air movements, the humidity, and the pressure of the atmosphere to forecast the weather. It takes hours to perform these calculations on supercomputers. Google just unveiled GenCast – a model that can produce results within eight minutes thanks to artificial intelligent.

An increasingly accurate model

The simulator analyzes 40 years’ worth of historical weather data to quickly predict the evolution of the weather over a period of 15 days. The rapidity of the simulator does not affect its accuracy or efficiency. The supercomputer was tested on several recent climatic phenomena and compared to the European ENS Model, which is the standard reference, it produced better forecasts for 15 days in 97% cases.

The supercomputer could forecast extreme conditions of weather, including storms.

Google’s model, for instance, could have given warnings on tropical storms 12 hours in advance. The study found that it was able to provide a better estimate of the storm track and the heavy rain, giving more time for warning and protecting populations.

Although the results of this model are encouraging, it should not be used to replace existing models. In fact, the artificial intelligence must be fed data from existing models and their various field observation station. The fact is that meteorologists are not yet confident in artificial intelligence, as it seems to them like a bit of magic, or nebulous. The meteorologists prefer physical simulations, as they are more rigourous and concrete. Both approaches are likely to be employed in parallel. Google’s code will actually be made available as an open-source project, so that all meteorologists can test it and form their own opinions.

プロ Python自動化サービス カスタムソリューション