Google DeepMind Unveils WeatherNext 2: Revolutionary AI Model Transforms Weather Forecasting with 8x Speed Boost

DeepMind releases a new weather forecasting model for more accurate predictions

Google's DeepMind , a new version of its AI weather prediction model. The company promises that it "delivers more efficient, more accurate and higher-resolution global weather predictions." To that end, it should be able to provide accurate forecasts up to two weeks out, including information on temperature, pressure and wind. It should also be able to better predict tropical storm tracks, according to researchers. This means that predictions of a hurricane's path should be accurate up to three days out. The previous model only predicted things up to two days ahead of the storm. The model also brings hourly forecasts into the mix. All told, Google says this new model is eight times faster than the previous iteration. This should help businesses like energy traders make more precise decisions, . "It gives you a more granular forecast," DeepMind AI researcher Akib Uddin said. "Many other industries are quite interested in these one-hour steps. It helps them make more precise decisions. Their goal is, how can they make their business more resilient to weather?" The improvements here stem from a new approach to weather models, as explained in this . Older methods demanded the use of machine learning models that were built for image and video generation. This required repeated processing to ensure an accurate result. The new model only requires a single processing step, which also reduces reliance on costly AI computing systems. AI may not be , despite what proponents want you to believe, but it is . Newer models typically , even those that rely on supercomputers. These models aren't perfect, however, as even Google has acknowledged that WeatherNext 2 will likely struggle to predict outlier rain and snow events. This is due to gaps in the training data. "It's one limitation of our forecast, but one that we are working on improving," DeepMind research scientist Ferran Alet told Bloomberg.

Google apps are getting better weather forecasts

Google's WeatherNext 2 is here, and it's not just some advanced AI model to help with Gemini. It is designed to give faster and much more detailed global weather predictions. This should fundamentally change how Google generates forecasts, making them smarter, quicker, and more useful. The speed improvements alone are hard to believe. The biggest problem with old physics-based models was that they often took hours on a supercomputer to produce comparable data. Now, that massive time sink is almost entirely eliminated. WeatherNext 2 is capable of generating forecasts eight times faster than its predecessor. This lets you get updated predictions much more quickly than before, which didn't take very long to begin with. The real magic happens when you look at how well it operates. This AI model can generate hundreds of possible weather scenarios in under a minute, and it does this using just a single Tensor Processing Unit, which is also hard to believe. The numbers are just as impressive when talking about accuracy. Google reports that WeatherNext 2 surpasses the previous state-of-the-art WeatherNext model on a whopping 99.9% of variables and lead times. This covers everything you care about, like temperature, wind, and humidity, and forecasts spanning from zero up to 15 days out. This new model can also give higher-resolution predictions, drilling down to one-hour increments. This level of detail is exactly what you need when planning a trip or deciding when to run errands, not just a vague idea of the weather for the whole afternoon. One of the most interesting parts of WeatherNext 2 is its ability to predict hundreds of possible weather outcomes from a single starting point. This includes those low-probability, but potentially catastrophic, weather events that are the most important to plan for, so it is ready to move forward should the weather change for the worse. This is powered by a new AI modeling technique called a Functional Generative Network, or FGN. This technology is clever because it injects "noise" directly into the model's architecture. This makes sure that even though the model is creating many different possible outcomes, the forecasts it generates remain physically realistic and interconnected. It's like creating hundreds of slightly different, but still plausible, timelines for the weather. You're going to see more accurate and detailed forecasts across a huge range of apps. This includes updates to forecasts in Search, Gemini, and the dedicated Pixel Weather app. The improved weather information is also hitting the Google Maps Platform's Weather API. In the coming weeks, it will also help power the standard weather information you see directly in Google Maps. Beyond regular apps, Google is also making this powerful data accessible to the scientific and business communities. So if you are interested, the forecast data from WeatherNext 2 is now available in Earth Engine and BigQuery. This isn't the end of the upgrades coming to our weather apps. Google DeepMind and Google Research will continue to research more capabilities to improve the models even further. This includes integrating new data sources and expanding access even further down the road. However, now we can benefit by seeing our own apps become faster and more accurate.

WeatherNext 2 is Google's most accurate forecasting model, now used by Pixel Weather & Search

Google DeepMind and Google Research today announced WeatherNext 2 as its "most advanced and efficient forecasting model." Notably, it's helping power forecasts in Google's consumer apps, including Pixel Weather. At a high-level, "WeatherNext 2 can generate forecasts 8x faster and with resolution up to 1-hour." It can predict wind speed and direction, precipitation, pressure, and other weather variables. This model's main advancement is predicting "hundreds of possible weather outcomes from a single starting point." This includes "low-probability, but catastrophic, weather events." Similar predictions would previously take "take hours on a supercomputer using physics-based models," but can now be done in under a minute using a single TPU and this AI approach. Under-the-hood, WeatherNext 2 generates four 6-hour forecasts per day. The model's input is the "most recent global weather state." A Functional Generative Network (FGN) is then used to produce slightly different forecasts, with those generated predictions then fed back into the model. This approach is particularly useful for predicting what meteorologists refer to as "marginals" and "joints." Marginals are individual, standalone weather elements: the precise temperature at a specific location, the wind speed at a certain altitude or the humidity. What's novel about our approach is that the model is only trained on these marginals. Yet, from that training, it learns to skillfully forecast 'joints' -- large, complex, interconnected systems that depend on how all those individual pieces fit together. This 'joint' forecasting is required for our most useful predictions, such as identifying entire regions affected by high heat, or expected power output across a wind farm. WeatherNext 2 surpasses Google's previous model "on 99.9% of variables (e.g. temperature, wind, humidity) and lead times (0-15 days)." The company has now incorporated WeatherNext technology and models "into the core forecasting system that powers all of Google's weather features." As such, you will see more accurate weather forecasts in Search, Gemini, Pixel Weather, and soon Google Maps. Meanwhile, businesses, scientists, and developers can access WeatherNext 2 via Google Cloud Vertex AI, Big Query, and Earth Engine. Looking ahead, Google will be "integrating new data sources, and expanding access even further."