Google is stepping up its game in digital weather forecasting with "WeatherNext 2," a high-performance forecasting model capable of predicting hourly weather changes. This new generation of AI, built on DeepMind’s work in atmospheric modeling, is now being used by several flagship services within the Google ecosystem. The search engine, the Gemini Assistant, the Google Pixel weather app, and soon Google Maps all rely on WeatherNext 2 to provide more detailed and reliable forecasts.
This widespread adoption reflects a profound transformation in the meteorological sector. While traditional forecasting models relied on complex physical simulations that were computationally intensive, AI introduces a new approach that is more flexible and capable of directly learning atmospheric dynamics. In a context marked by more frequent and intense weather events, the ability to forecast on an hourly basis with increased accuracy opens up significant opportunities for public services, businesses, and citizens.
WeatherNext 2, a highly accurate model developed by Google DeepMind
WeatherNext 2 leverages recent advances in neural atmospheric forecasting models. Thanks to an optimized architecture, the AI analyzes billions of data points from satellites, radars, ground stations, and global climate models. Compared to traditional numerical systems, which are often limited by the computational power required to solve physical equations, WeatherNext 2 generates faster forecasts that are, in many cases, more accurate.
Initial internal evaluations show that the model significantly reduces the margin of error in forecasting temperatures, wind, precipitation, and cloud formations. Its accuracy is particularly notable for localized and rapid weather events, which are traditionally difficult to predict using conventional models.
How WeatherNext 2 manages to forecast hourly changes
The strength of WeatherNext 2 lies in its ability to capture complex phenomena using deep learning. The model:
- simultaneously incorporates visual data (satellite images, radar fields),
- interprets nonlinear atmospheric dynamics,
- learns about cloud formation and dissipation patterns,
- models urban microclimate phenomena,
- generates detailed, high-resolution forecasts covering several hours.
Thanks to this approach, WeatherNext 2 is able to forecast rapid changes—such as isolated showers, gusts of wind, fog patches, or local thunderstorms—much more accurately than traditional numerical weather prediction models.
Practical applications for citizens and strategic sectors
Hour-by-hour forecasts offer significant opportunities in sectors where the weather directly affects business operations. Among the areas most impacted are:
- aviation, with more accurate forecasts of turbulence and flight conditions,
- agriculture, through localized monitoring of precipitation and humidity,
- energy, to optimize renewable energy production and adjust demand in real time,
- risk management, particularly during periods of extreme weather,
- logistics and urban mobility, with improved traffic forecasting and visibility,
- event planning, where every hour counts for the organizers.
WeatherNext 2 thus enhances the ability to make operational decisions based on reliable, up-to-date data.
Access, availability, cost, and timeline: a gradual ramp-up
WeatherNext 2 is already integrated into Google products in several regions, including North America, Western Europe, and the Asia-Pacific region. Users can access it automatically through Google Search, Gemini, and the Pixel Weather app, without any additional setup. Integration with Google Maps will be rolled out gradually as regional updates are released.
The model is available at no additional cost to users, as Google has chosen to integrate WeatherNext 2 into its existing services. At the same time, a dedicated API for weather partners and public agencies is currently being rolled out. Updates to the international rollout schedule are published on the official Google Research and Google DeepMind blogs, which serve as a reference for tracking the expansion of WeatherNext 21.
Scientific uncertainties remain
Despite its impressive performance, WeatherNext 2 still has some limitations:
- AI relies heavily on the quality of the input data,
- Weather extremes remain difficult to predict with certainty,
- the model may lack scientific interpretability in certain scenarios,
- The transition between AI models and physical models still requires cross-validation.
These limitations serve as a reminder that AI is a valuable tool, but it does not completely replace traditional approaches.
Ethical and Geopolitical Implications of Weather Control by Big Tech
The arrival of an advanced weather model developed by a private company as large as Google raises several questions. The concentration of predictive capabilities in the hands of a few companies could pose issues regarding scientific sovereignty and equitable access to information. Public authorities are also raising concerns about the transparency of these models and their compliance with emerging regulations, particularly the European AI Act2.
Finally, the management of meteorological data—which influences agriculture, energy, transportation, and civil protection—is becoming a strategic issue on a global scale. The governance of these systems will therefore need to be carefully regulated.
Toward AI-Enhanced Meteorology
WeatherNext 2 marks a decisive milestone in the evolution of weather forecasting. By providing hourly accuracy, AI is transforming the way citizens, businesses, and public services understand weather patterns. This breakthrough reflects a major trend toward hybrid climate modeling, where neural networks and physical simulations converge to offer a more detailed and actionable understanding of atmospheric phenomena.
Meteorology is thus entering a new era, in which artificial intelligence is becoming an essential tool for climate resilience, infrastructure management, and public safety.
Learn more
To understand how AI is already transforming scientific modeling and predictive analytics, check out: DINOv3 by Meta, self-supervised learning for high-precision visual analysis
References
1. Google DeepMind. (2025). WeatherNext 2 Technical Overview and Deployment Notes.
https://deepmind.google
2. European Commission. (2025). AI Act, Guidelines for AI Systems in Critical Infrastructures.
https://ec.europa.eu
