Google released version 3 of its WeatherNext AI weather forecasting model on September 8, with the biggest change being the addition of satellite weather data ingestion. The update, detailed in a white paper, shortens the lag time between current weather conditions and generating a new forecast.
AI weather models have become major players in forecasting, offering performance similar to traditional physics-based models while requiring far less computing power to run. That efficiency means they can be run more frequently, producing updated forecasts at higher resolution and shorter intervals than traditional numerical weather prediction systems.
Most AI weather models have relied entirely on reanalyses — comprehensive global snapshots of the atmosphere produced every six to twelve hours that combine all available weather data into a single consistent picture. While useful, reanalyses can lose information from raw data sources during the blending process, and the six-to-twelve-hour production cycle means they always lag behind current conditions. WeatherNext 3 now ingests some satellite data directly, bypassing the reanalysis step and capturing more current conditions for improved short-term accuracy.
The update represents a broader trend of AI weather models becoming more sophisticated in how they consume data, moving beyond training solely on reanalysis products toward incorporating real-time observations. This evolution could lead to meaningfully better short-term forecasts, particularly for rapidly changing weather events where every hour of lag matters for emergency preparedness and daily planning.