WeatherNext 3: More accurate, timely, and local weather forecasts

Summary
This video presents WeatherNext 3, an AI-powered global weather forecasting model developed by Google DeepMind. The system is introduced and explained by Product Lead Nofar Peled Levi and Research Scientist Stephan Rasp, who outline how it improves spatial resolution, update frequency, and practical utility for industries such as renewable energy.

What is shown

• [00:15] Traditional vs. AI forecasting overview: Archive footage and graphic diagrams contrasting physics-based numerical simulations with AI trained on historical weather observations.
• [01:05] Forecasting cadence and resolution graphics: Visualizations showing hourly forecast generation across the globe at up to 5 km resolution.
• [01:36] Data assimilation: Animations highlighting the ingestion of raw satellite imagery and ground weather station data directly into the model rather than relying solely on 6-hour reanalysis cycles.
• [02:18] Spatial resolution demonstration: A whiteboard-style explanation showing grid sizing (such as a 25 km × 25 km block) and comparing low-resolution regional averaging to finer local detail.
• [03:06] Multi-resolution outputs: Visual map renders displaying three native resolutions produced in a single pass (25 km for general atmosphere, 9 km for wind and pressure, and 5 km for temperature and humidity).
• [03:27] Renewable energy indicators: Visual models highlighting 100-meter altitude wind speed and direction data for wind turbines, as well as cloud cover and solar radiation tracking for photovoltaic arrays.
• [04:08] Product integration: Explanations of how WeatherNext 3 forecasts integrate across Google products like Search, Gemini, and Google Maps.

Claims & numbers

• The presenters claim WeatherNext 3 is the first global operational AI weather model to produce a new forecast every hour of the day (compared to traditional 6-hour reanalysis refresh intervals).
• Stephan Rasp states the model achieves up to 5 km spatial resolution with hourly time steps and state-of-the-art skill across most atmospheric variables.
• Nofar Peled Levi states the model provides three native resolutions in a single pass: 25 km for broader atmospheric changes, 9 km for surface variables like wind and pressure, and up to 5 km for temperature and humidity.
• Stephan Rasp states WeatherNext 3 specifically predicts wind speed and direction at 100 meters altitude (the typical hub height of modern wind turbines), as well as cloud cover and solar radiation for solar farms.
• Rasp states the model will be deployed across Google surfaces including Google Search, Gemini, and Google Maps to reach billions of users.

Notable quotes

• [01:04] "WeatherNext 3 is the first global operational weather model that produces a new forecast every hour of the day with up to 5-kilometer resolution..." — Stephan Rasp
• [01:52] "The atmosphere does not move in a six-hour leaps, right? ... Changes can happen in minutes." — Nofar Peled Levi
• [03:02] "WeatherNext 3 actually provide with three native resolutions in single pass." — Nofar Peled Levi

Assessment
This is an official Google DeepMind product overview and explainer video showcasing WeatherNext 3. The technical outputs are presented through pre-rendered maps, motion graphics, and illustrative data visualizations rather than an interactive software demo or raw benchmark comparison tables.

Described by gemini-3.8-flash on 2026-10-06 from the video's audio and frames.