Artificial intelligence in weather predictions is not entirely new. However, Google DeepMind has just announced GenCast, a model that promises to revolutionize weather forecasting.
Please follow us on Facebook and Twitter.
According to Google, GenCast is a significant improvement over previous AI systems used for weather predictions. Unlike earlier models, each forecast generated by GenCast consists of 50 or more predictions, representing different meteorological scenarios.
Previously, AI tools were limited to deterministic models, meaning they could only generate a single weather forecast that they believed was the most accurate. With GenCast, Google aims to raise the standard for both daily forecasts and extreme weather event predictions.
In both short-term and up to 15-day forecasts, Google claims that GenCast outperforms the ENS system from the European Centre for Medium-Range Weather Forecasts, which is considered the best in the world. This is no small claim.
Google tested GenCast against the ENS using the 2019 climate forecast, having trained the model with data up to 2018. In the 1,320 combinations of variables and timeframes they tested, GenCast was more accurate than the ENS system in 97.2% of the cases.
GenCast, Google’s AI, is set to revolutionize weather forecasting.
One of the most interesting examples Google shared about GenCast was its prediction of Typhoon Hagibis, which struck Japan in October 2019. The AI was able to forecast a range of possible paths up to 7 days in advance, with a high degree of certainty.
Google DeepMind’s new AI was trained using decades of data from the ERA5 archive of the European Centre for Medium-Range Weather Forecasts. On the technical side, GenCast requires just one tensor processing unit (TPU) on Google Cloud servers to generate a 15-day weather forecast. This means the system could be scaled to much larger levels using dedicated infrastructure.
Google plans to release both the code and the weights for GenCast, along with its weather forecasts, to encourage adoption and usage. This could help the technology become a valuable addition to existing weather prediction systems, enabling more accurate estimates and better early warning systems for storms and extreme events.
As mentioned earlier, Google has expanded its use of AI in weather forecasting. A few years ago, the company introduced a tool that could predict whether it would rain in the next 90 minutes. In 2023, it unveiled a model capable of predicting hurricanes with greater accuracy.
