Google’s DeepMind team introduced an AI model named GenCast for predicting weather this week.
In a paper published in Nature, DeepMind researchers revealed that GenCast surpasses the European Centre for Medium-Range Weather Forecasts’ ENS — reportedly the leading operational forecasting system worldwide.
In a blog post, the DeepMind team provided a more straightforward explanation of the technology: While their previous weather model offered a single, best estimate of future weather, GenCast generates an ensemble of 50 or more predictions, each depicting a potential weather trajectory, resulting in a complex probability distribution of future weather scenarios.
In terms of performance compared to ENS, the team stated that they trained GenCast using weather data up to 2018 and tested its forecasts for 2019, determining that GenCast was more accurate 97.2% of the time.
Google mentions that GenCast is included in its collection of AI-driven weather models, which are being integrated into Google Search and Maps. They also intend to release real-time and historical forecasts generated by GenCast, available for use in individual research and models.
