The organization claims its AI weather forecasting model is quicker and more precise than ever.
The organization claims its AI weather forecasting model is quicker and more precise than ever.
Google unveiled an enhanced AI weather forecasting model today that will be integrated into popular applications such as Search, Gemini, and Pixel smartphones.
The company’s AI-driven predictions have demonstrated notable accuracy to date. AI weather systems have further shown they can deliver faster and more efficiently than traditional physics-based methods. Up to this point, Google’s endeavors in this field have been largely experimental. Now, it is positioning those predictions as a marketable advantage for Google offerings.
“We’re taking it out of the lab and genuinely placing it in the hands of users more extensively than before and shedding the experimental label since we are confident that our forecasts are extremely effective and beneficial,” said Peter Battaglia, senior director of research and sustainability at Google DeepMind, during a briefing with journalists.
The latest AI model, WeatherNext 2, can produce forecasts eight times quicker than Google’s earlier model and shows improved accuracy in predicting 99.9 percent of variables like temperature and wind. WeatherNext 2 can generate numerous possible outcomes from a specific starting point. It takes less than a minute utilizing one of Google’s TPU chips to yield a prediction, which the company claims would normally take hours using physics-based models on a supercomputer.
Those traditional models are demanding in computation as they essentially attempt to replicate the complex physics of the atmosphere to generate forecasts. AI systems, however, seek to identify patterns from historical weather data to anticipate future conditions.
Google streamlined its methods by utilizing a technique it terms a Functional Generative Network (FGN) in WeatherNext 2. Previous AI weather models required extensive processing to generate a single forecast. FGN is more productive because it integrates noise—or targeted randomness—into the model with each input, allowing WeatherNext 2 to produce various possible outcomes in a single operation.
These advancements enable WeatherNext 2 to predict up to 15 days in advance and create hourly forecasts. Google is betting that this feature will appeal to corporate clients as well as individual users.
“We’ve noticed that sectors such as energy, agriculture, transportation, logistics, and many other markets are quite interested in these hourly increments. It aids them in making more exact choices that impact their operations,” stated Akib Uddin, a product manager at Google Research, during the call.
In addition to integrating WeatherNext 2 into Maps, Search, Gemini, and Pixel Weather, Google is also launching an early access initiative for clients interested in custom modeling. The forecasting data is also accessible in Google Earth Engine for geospatial studies and BigQuery for extensive data analysis.
Google faces competition in the quest to leverage generative AI for weather forecasts. The European Center for Medium-Range Weather Forecasts, Nvidia, Huawei, and several other firms are similarly developing their own AI weather systems.
