Google has introduced an updated weather forecasting model, WeatherNext 2, which will become part of the company's products such as Search, Gemini, Maps, and Pixel smartphones.
Google says WeatherNext 2 is eight times faster than its predecessor and is 99.9% more accurate at predicting key weather variables, including temperature and wind speed. The model can generate hundreds of possible scenarios in less than a minute on Google's TPU chips, compared to hours of supercomputer time for traditional physical models.
WeatherNext 2 uses a new Functional Generative Network (FGN) architecture that allows for single-pass forecasting instead of multiple iterations required in previous AI models. The model forecasts weather up to 15 days in advance and generates hourly data.
Google says the more accurate hourly forecasts will benefit business customers in energy, agriculture, logistics, and transportation. WeatherNext 2 will also be available through Google Earth Engine and BigQuery for geospatial and large-scale analytics. The company is also launching an early access program for custom models for enterprise customers.
Google continues to compete with other companies developing AI forecasting, including NVIDIA, Huawei, and the European Centre for Medium-Range Weather Forecasts.