WeatherNext 2: Our most advanced weather forecasting model

The WeatherNext team has introduced WeatherNext 2, an AI forecasting model that delivers faster, higher-resolution global forecasts. The model can generate hundreds of possible weather scenarios from a single input in under a minute on one TPU, runs up to eight times faster than prior approaches, and produces predictions with up to one-hour resolution.

It outperforms the previous WeatherNext model on 99.9% of variables and lead times (0–15 days). Forecast data from WeatherNext 2 is now available in Earth Engine and BigQuery, and an early access program for custom model inference is open on Vertex AI. The technology has been used to upgrade weather information in Search, Gemini, Pixel Weather and Maps Platform, and it will soon help power weather details in Maps.

The model creates coherent ensemble variability by using independently trained neural networks and injecting noise in function space.

weathernext 2, weathernext, ai forecasting, global forecasts, high resolution, hourly resolution, tpu, earth engine, bigquery, vertex ai