How to Explore WeatherNext 3 in Google Weather Lab

Google Weather Lab lets you inspect live WeatherNext 3 output without a developer account. Open the map, switch layers, and compare the hourly ensemble mean against earlier models or traditional baselines.

WeatherNext 3, released by Google DeepMind and Google Research in September 2026, initializes every hour using live geostationary satellite mosaics. It produces 15-day forecasts for the main cycles and 48-hour runs for the interim hours. Key surface variables such as temperature and moisture appear at roughly 5 km resolution in some views, while other surface fields sit near 10 km and atmospheric variables near 25 km.

Weather Lab is an experimental research platform. Google states these are not official forecasts or warnings. Check your local meteorological agency for decisions that affect safety.

Open Weather Lab and locate WeatherNext 3

1. Go to the Weather Lab site from the Google DeepMind WeatherNext page or the developers.google.com/weathernext guide.
2. Accept the experimental research notice if one appears.
3. Use the map controls to pan and zoom to your region.
4. Select the WeatherNext 3 layer. The interface lists it alongside WeatherNext 2 and Global MetNet.

The WeatherNext 3 layer shows ensemble mean values for 2 m temperature, total precipitation, 10 m wind speed, and sea level pressure. Time steps are hourly. The main 00, 06, 12, and 18 UTC cycles extend to 15 days; other hourly inits reach about 48 hours.

You can step forward and backward through the forecast timeline with the time slider. Watch how precipitation bands move between successive hourly runs.

Compare models on the same map

Weather Lab places WeatherNext 3 next to WeatherNext 2 (FGN) and Global MetNet. Switch layers to see differences in timing and spatial detail.

• WeatherNext 3 uses 1-hour steps and displays the mean of a 64-member ensemble.
• WeatherNext 2 uses 6-hour steps on a coarser grid.
• Global MetNet focuses on precipitation at 15-minute steps out to 12 hours.

Zoom into a coastal or mountainous area. The higher-resolution surface fields in WeatherNext 3 resolve valleys and shorelines more clearly than the 25 km WeatherNext 2 grid.

Track tropical cyclones

Weather Lab also tracks tropical cyclones with WeatherNext Cyclones. Predicted tracks appear alongside ECMWF ENS and HRES baselines. Observed tracks come from NHC/JTWC TCVitals in real time and IBTrACS for history.

1. Open the cyclone section of the map.
2. Select an active or recent storm.
3. Toggle ensemble member tracks to see the spread of possible paths.
4. Open deep-dive charts for intensity and position over time.

You can download experimental cyclone track data in CSV or ATCF format. Real-time data follows Google’s experimental terms. Historical data older than one hour is under Creative Commons Attribution 4.0.

This cyclone view is separate from the global WeatherNext 3 surface layers. Use both when a storm is approaching land.

Practical limits and best use

Weather Lab shows research output. Ensemble means smooth individual member extremes. Precipitation skill improves relative to earlier models, yet local radar and official warnings remain the primary source for short-term hazards.

For everyday planning, the same model already feeds Google Search, the Gemini app, and Google Maps. Open those products, note the update time, and recheck an hour later because new satellite data arrives hourly. For bulk data, see our guide to accessing WeatherNext 3 forecasts on Google Cloud.

Tips for clearer views

• Refresh the page after each new hourly initialization if the timeline feels stale.
• Use the layer legend to confirm units (temperature in °C or °F depending on locale, precipitation in mm).
• Compare a 6-hourly WeatherNext 3 cycle against the same hour from WeatherNext 2 to isolate the effect of the satellite input.
• Download cyclone files only when you need them for analysis; the map is sufficient for most visual checks.

Conclusion

Weather Lab gives a direct look at WeatherNext 3’s hourly ensemble fields and cyclone tracks. Open the map, select the layer, step through time, and compare against earlier models. Treat every view as experimental research and confirm critical decisions with your local weather service.

Sources

• Weather Lab — Google for Developers
• Introducing WeatherNext 3 — Google Blog, 3 September 2026
• WeatherNext 3 model guide — Google for Developers
• WeatherNext 3: More accurate, timely, and local weather forecasts — Google DeepMind