Google Launches WeatherNext 3 With Hourly Forecasts at Up to 5km Resolution

Google DeepMind and Google Research have introduced WeatherNext 3, which the company describes as its most advanced and accurate global weather model to date. The system generates a new forecast every hour at resolutions down to 5 kilometers and has already begun powering weather experiences across Google products.



Key surface variables such as temperature and moisture are forecast at 5-kilometer resolution, other surface variables at 10 kilometers, and atmospheric variables such as wind speed at 25 kilometers. Google says the result is a weather picture roughly five times sharper than WeatherNext 2, which produced forecasts on a 25-kilometer grid in six-hour increments.


A major advance comes from incorporating live observations. Most AI weather models, including WeatherNext 2, are trained on data from numerical weather prediction systems, the supercomputer-driven physics simulations used in traditional forecasting. Google's new system ingests hourly mosaics of live global geostationary satellite data alongside traditional historical analysis, allowing each forecast to incorporate the latest satellite observations available.

The model also trains directly on sparse weather station observation data, helping its forecasts account for regional details such as topography. Google says the approach is particularly useful across Latin America, Africa, and Asia-Pacific, regions historically underserved by high-resolution forecasting because of the computing costs associated with traditional regional models.

Global weather models have historically struggled with rain and snow because precipitation systems can develop rapidly at scales that are difficult to model. WeatherNext 3 was trained using NASA's satellite-based Integrated Multi-satellite Retrievals for GPM dataset, known as IMERG, along with Google's own global precipitation reanalysis based on satellite radar. In medium-range evaluations, Google reports Continuous Ranked Probability Score improvements of up to 60 percent against IMERG, 30 percent against MRMS, and 10 percent against rain gauge measurements at early lead times. For forecasts a day or more ahead, the company says users will see precipitation predictions that are up to 50 percent more accurate, with the largest improvements in regions where forecasts have historically been less reliable.

For renewable energy production, the system predicts 100-meter wind speeds at roughly turbine height, along with high-resolution cloud cover and solar radiation levels, helping grid operators and renewable energy developers estimate wind and solar output and match generation with demand.


WeatherNext 3 is now beginning to power weather experiences in Google Search, the Gemini app, Google Maps, the Google Maps Platform Weather API, and Google Earth Engine. Researchers, developers, and businesses can also query hourly forecast data through BigQuery and Earth Engine or bulk-download it from Google Cloud Storage.