Scientists at Google DeepMind and Google Research have developed a new AI model, WeatherNext 3, that predicts weather with unprecedented accuracy. Unlike traditional meteorological models, which rely on complex mathematical equations, WeatherNext 3 uses deep learning techniques to forecast everything from temperature to humidity with greater precision.

This isn't just a tech update: it could mean more accurate and frequent weather warnings, helping users avoid the surprise downpour. According to Samier Merchant, a Google senior staff engineer, the new model will be integrated into a range of Google products, from search to Maps. The improvements are significant, with WeatherNext 3 outperforming other deep-learning models and even traditional forecasts from the U.S. National Weather Service.

While the model's accuracy is impressive, it also faces challenges, such as predicting rain and handling a wide area, which it addresses by operating at a 5 km resolution. The model can now produce hourly forecasts and has been trained to predict specific weather data stations, offering granular predictions. This, combined with the ability to ingest real-time satellite data, promises faster and more accurate forecasts.

Google's push to directly incorporate raw observations for high-resolution global forecasts could revolutionize weather forecasting, making it more accessible and affordable. While other AI weather startups like WindBorne have been doing this for a while, Google's model is still leading the pack. The potential benefits are enormous, from improved crop yields in developing countries to more reliable renewable energy projects.