Google's AI research team, DeepMind, released the latest version of its AI weather forecasting model earlier (November 18)."WeatherNext 2"The official statement emphasizes that the new version has improvements in efficiency, accuracy, and resolution. It can provide global weather forecasts up to two weeks in length (including temperature, air pressure, and wind speed), and its processing speed is eight times that of the previous generation model, helping energy traders and other industries make more accurate decisions.
Key upgrades: hourly forecasts, 3-day storm path
The key upgrade of WeatherNext 2 is that it provides more granular forecasting capabilities.
First, the new model incorporates an "hourly forecasts" feature. DeepMind AI researcher Akib Uddin points out that many industries are interested in this hourly forecasting step, which helps strengthen their businesses' resilience to weather changes.
Secondly, in terms of disaster prediction, WeatherNext 2's ability to accurately predict the path of tropical storms (such as hurricanes) has been improved from 2 days in advance to 3 days in advance compared to the previous generation.
Core technology: The new model adopts "single-processing" to reduce computational dependence.
The significant improvement in effectiveness stems from a new approach to meteorological models. According to newly published...Research paperThe old method required the use of machine learning models specifically designed for image and video generation, which resulted in repeated processing to ensure accurate results.
WeatherNext 2's new model requires only a "single processing step," thus significantly reducing its reliance on expensive AI computing systems.
Limitations: Extreme weather remains a challenge; NVIDIA and Microsoft have both taken precautions.
Although AI typically outperforms traditional supercomputers in weather forecasting, Google acknowledges that WeatherNext 2 still has limitations.
DeepMind research scientist Ferran Alet displayDue to gaps in the training data, the new model may still encounter difficulties in predicting outlier rain and snow events.
Currently, AI-powered weather forecasting has become a fiercely contested area, including... Companies such as NVIDIA, Microsoft, AccuWeather, and HuaweiThey have all invested in this field one after another.
