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.
Advertisement – scroll for more content
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.”
FTC: We use income earning auto affiliate links. More.