- WeatherNext 3 offers hourly forecasts with a resolution of up to 5 km, five times more accurate than its predecessor.
- The model is trained using real-time satellite data and weather stations, improving rainfall accuracy by up to 60%.
- It includes specific predictions for renewable energy, such as wind speed at 100 meters and solar radiation.
- It is now available on Google Search, Gemini, Google Maps and Google Cloud, with particular benefit for underserved regions.
Google has officially unveiled WeatherNext 3, its new artificial intelligence model for weather forecasting, which promises to be the most advanced and accurate to date. Developed by Google DeepMind and Google Research, this system is already available in products such as Search, Gemini, Google Maps, and Google Cloud, and represents a significant leap forward in weather prediction, offering greater accuracy thanks to AI on a global scale.
The main innovation of WeatherNext 3 lies in its training method. Unlike other AI weather models, which rely on supercomputer simulations with a six-hour lag, this new system learns directly from real-time observations. It does this by ingesting a mosaic of geostationary satellite data and data from dispersed weather stations, allowing it to generate hourly forecasts with a resolution of up to 5 kilometers.
A leap in resolution and frequency
Resolution is one of the most significant improvements. While WeatherNext 2 used a 25-kilometer grid and updated every six hours, WeatherNext 3 offers a global image approximately five times sharper. This translates into a much greater ability to detect rapidly developing local phenomena such as storms, fronts, or precipitation systems—crucial for an effective response to severe weather events.
Furthermore, the model distinguishes between different variables: temperature and humidity at 5 kilometers, surface variables at 10 kilometers, and wind speed at 25 kilometers. This granularity is especially relevant for communities near coasts, valleys, or mountain ranges, where conditions can vary drastically within just a few kilometers. Traditional models, lacking this level of detail, often overlook local extremes that are now captured.
Accuracy in precipitation
One of the biggest challenges in global meteorology has always been predicting rain and snow. Physics-based systems struggle to model rapid cloud processes at small scales, resulting in fuzzy estimates or errors at the boundaries of severe storms. To address this, WeatherNext 3 is trained using two high-quality data sources: NASA's IMERG system and a proprietary global reanalysis based on satellite radar.
The results are compelling. In medium-range forecast assessments, the model shows up to a 60% improvement in CRPS scores compared to IMERG, 30% compared to MRMS, and 10% compared to earlier-range rain gauge measurements. This means that Google Search and Gemini users will see precipitation forecasts that are up to 50% more accurate, helping them know when it's going to rain, especially in regions where forecasts have historically been less reliable.
Benefits for clean energy
Another major contribution of WeatherNext 3 is its focus on renewable energy production. The model includes specific wind speed predictions at 100 meters, approximately the height of wind turbines, as well as high-resolution cloud cover and solar radiation data. This information is vital for electricity grid operators and solar and wind farm developers to accurately estimate how much energy their facilities will generate and adjust supply to meet demand.
This advancement is particularly important for regions in Latin America, Africa, and Asia-Pacific, which have historically been underserved by high-resolution models due to the enormous computational cost of traditional regional systems. With WeatherNext 3, billions of people and local businesses in these areas will be able to access detailed and reliable forecasts without needing their own supercomputer infrastructure.
Availability in Google products
Since its launch, WeatherNext 3 has been integrated into Google's main products. In Search and Gemini, users will notice a substantial improvement in long-range forecasts, with greater accuracy in planning outdoor activities. In Google Maps, weather information will be updated more frequently and in greater detail, while Google Maps Platform and Google Cloud will offer this data to developers and businesses for their own applications.
The combination of 5-kilometer resolution, hourly updates, and training with real-world data makes WeatherNext 3 a revolutionary tool. It not only enhances the end-user experience but also opens up new possibilities in sectors such as agriculture, logistics, and emergency management, where having accurate and up-to-date weather widgets can mean the difference between prevention and disaster.
With this launch, Google reinforces its commitment to artificial intelligence applied to real-world problems, demonstrating that the combination of satellite data, machine learning, and cloud computing can offer tangible and scalable solutions on a global scale. Meteorology, one of the most complex sciences with the greatest impact on daily life, thus takes a giant leap toward a future where weather forecasting will be increasingly reliable and accessible to everyone.
