Google Search, the Weather app, and Gemini now offer more accurate forecasts. They use the new WeatherNext 2 model

The Google DeepMind and Google Research teams jointly unveiled WeatherNext 2 on Monday, their most advanced artificial intelligence model for weather forecasting to date. The new technology reportedly can generate forecasts 8 times faster than the previous version and offers temporal resolution down to individual hours. Google is already deploying the model into its services, including Search, the Gemini assistant, and the mobile Weather app (Pixel Weather).

Extremely fast forecasts for various scenarios

The biggest advantage of WeatherNext 2 is its ability to create hundreds of possible weather development scenarios from a single starting point. While traditional models compute for hours on supercomputers, Google’s new model can handle each forecast in less than a minute on a single TPU processor. This allows meteorologists to better prepare for extreme situations and plan for the worst-case scenarios.

The model uses a new technology called Functional Generative Network (FGN), which adds controlled “noise” to the model’s architecture. This ensures that forecasts remain physically realistic and interconnected – for example, when the model predicts a high temperature in a certain area, it also correctly estimates related changes in humidity and air pressure.

More accurate in all respects

According to the official blog post, WeatherNext 2 surpasses the previous model in 99.9% of monitored parameters such as temperature, wind, or air humidity. Forecasts are accurate for a horizon of 0 to 15 days ahead, with the model capable of working with hourly resolution. This means significantly more detailed information for planning daily activities.

Interestingly, the model learns only from individual meteorological elements – for example, temperature at a specific location or wind speed at a certain altitude. Nevertheless, it can predict complex systems such as entire areas affected by a heatwave or the expected output of wind farms.

Where can you find the new forecasts?

Google has already integrated WeatherNext technology into Search, the Gemini assistant, the Pixel Weather app, and the Google Maps Platform Weather API. In the coming weeks, forecasts based on the new model will also appear directly in the Google Maps app.

Practical impact on everyday life

For ordinary users, WeatherNext 2 primarily means more reliable forecasts when planning activities. The model can better estimate the risk of extreme phenomena, which helps not only individuals but also businesses in deciding on flight routes, supply chains, or planning outdoor events. Google has already tested the technology in experimental cyclone forecasts, where it helped meteorological agencies make decisions based on a wide range of possible scenarios.

The WeatherNext 2 model represents a significant advancement in meteorological forecasting. The combination of processing speed, accuracy, and the ability to model various scenarios makes it a tool that can fundamentally change how we work with weather forecasts.

Do you rely on forecasts from Google Search or Pixel Weather?

Source: Google