Google announced on November 17 that it would comprehensively upgrade its weather forecast service, adopting the newly developed AI weather model "WeatherNext 2". This new model will be integrated into many of Google's popular products, including search, Gemini and Pixel phones.

According to reports, Google’s AI-enhanced weather forecast has already performed well in terms of accuracy. Previously, AI weather models have surpassed traditional physics-based models in the speed and efficiency of generating forecasts. Google said WeatherNext 2 can use Google's TPU chip to complete forecasting tasks that would take traditional supercomputers hours to complete in less than a minute.

"We are putting laboratory results into the hands of users on a large scale. We believe these predictions are so efficient and practical that they are no longer limited to experimental nature." Peter Battaglia, senior director of research and sustainability at Google DeepMind, said at a press conference.

It is reported that the new generation AI model WeatherNext 2 is eight times faster than the previous generation in prediction speed, and it is also more accurate in predicting 99.9% of variables (such as temperature and wind). The model can generate hundreds of possible outcomes based on the starting point and can give predictions 15 days in advance, supporting hour-by-hour weather forecasts. This is particularly important for users in fields such as energy, agriculture, transportation and logistics, as they have higher demand for precise information at the hourly level.

The reason why WeatherNext 2 can achieve efficient forecasting is because it adopts a new strategy called "Functional Generative Network (FGN)". FGN generates multiple predictions in a single step by introducing noise (targeted randomness) at each input. This technique significantly reduces the need for repeated calculations.

Google said that WeatherNext 2 will not only be launched on products such as maps, search, Gemini and Pixel Weather, but will also provide early access plans for enterprise customers with customized needs. Relevant data will be open to research institutions and big data analysis customers through platforms such as Google Earth Engine and BigQuery.

Currently, competition in the field of weather AI is increasingly fierce, with institutions and companies including the European Center for Medium-Range Weather Forecasts, NVIDIA, and Huawei all developing their own AI weather models.