Google Unveils WeatherNext 2: A Breakthrough in Weather Forecasting

Google DeepMind and Google Research have launched WeatherNext 2, touted as the most advanced weather forecasting model to date. This new model is designed to generate forecasts eight times faster than its predecessor, with the capability to produce high-resolution predictions every hour. WeatherNext 2 can predict various weather variables, including wind speed, precipitation, and pressure, and is particularly adept at forecasting low-probability yet catastrophic weather events.

9To5GoogleGoogle DeepMind Unveils WeatherNext 2 - 8x Faster AI ForecastsEngadget

WeatherNext 2 utilizes a Functional Generative Network (FGN) to create hundreds of possible weather scenarios from a single input in under a minute. Traditional supercomputers would require hours to achieve similar results. The model generates four six-hour forecasts daily, relying on the most recent global weather data. Notably, it surpasses Google's previous model in 99.9% of variables, including temperature and humidity, across lead times of 0 to 15 days.

9To5GoogleGoogle DeepMind Unveils WeatherNext 2 - 8x Faster AI ForecastsEngadget

The model's ability to predict complex interconnected weather systems, referred to as "joints," is a significant advancement. By training solely on individual weather elements, WeatherNext 2 can forecast broader patterns, such as regional heat waves or expected power output from wind farms.

9To5Google

WeatherNext 2 is already integrated into various Google services, including Google Search, the Gemini assistant, Pixel Weather apps, and Google Maps. Businesses, scientists, and developers can access the model through Google Cloud Vertex AI, Big Query, and Earth Engine. This integration aims to enhance decision-making in industries reliant on accurate weather forecasts, such as energy trading and agriculture.

9To5Google

Despite its advancements, WeatherNext 2 is not without limitations. Critics point out that the model may struggle with predicting outlier rain and snow events due to gaps in its training data. Ferran Alet, a research scientist at DeepMind, acknowledged this shortcoming, stating, "It’s one limitation of our forecast, but one that we are working on improving."

Engadget

DeepMind researchers emphasize the model's potential to reshape industries by providing more granular forecasts. Akib Uddin, a DeepMind AI researcher, noted, "Many other industries are quite interested in these one-hour steps. It helps them make more precise decisions."

Engadget

Looking ahead, Google plans to expand WeatherNext 2's capabilities by integrating new data sources and broadening access to the model. This ongoing development aims to further enhance the accuracy and efficiency of weather predictions, solidifying WeatherNext 2's role in modern meteorology.

9To5GoogleGoogle DeepMind Unveils WeatherNext 2 - 8x Faster AI ForecastsEngadget

• “Weather predictions need to capture the full range of possibilities - including worst case scenarios, which are the most important to plan for,” — Google DeepMind WeatherNext Team

Google DeepMind Unveils WeatherNext 2 - 8x Faster AI Forecasts

• “It gives you a more granular forecast,” — Akib Uddin, DeepMind AI Researcher

Engadget

• “It’s one limitation of our forecast, but one that we are working on improving,” — Ferran Alet, DeepMind Research Scientist

Engadget