Google's WeatherNext 3 boosts forecast resolution to 5km, hourly across Search, Maps
A 5-kilometer grid now underlies Google's weather forecasts, replacing the 25-kilometer, 6-hour cadence of the prior WeatherNext 2 model. WeatherNext 3, built by Google DeepMind, updates hourly and folds in real-time satellite data, giving temperature and moisture readings at 5km resolution, other surface variables at 10km, and atmospheric variables such as wind speed at 25km. The model now powers forecasts across Google Search, the Gemini app, Google Maps, Google Maps Platform and Google Cloud, and developers can query the raw data through BigQuery, Earth Engine and Google Cloud Storage.
Weather targeting until now worked at the city or postal-code level, missing a rainstorm confined to one block while the next stays dry. Micro-climate precision opens the door to automated creative swaps: a coffee chain flipping from hot to iced drink ads as temperature spikes in a single 5km cell, or hardware and ride-share ads firing the moment wind and rain converge on one neighborhood. Google has left any advertising product unconfirmed. Agencies pitching weather-triggered creative today are extrapolating from a forecasting upgrade; a dedicated ad feature would need its own announcement.