谷歌DeepMind和Google Research推出了最新的AI天气预报模型WeatherNext 31。该模型在Operational WeatherBench测试中展现出最强的准确性，超越了微软、英伟达、欧洲中期天气预报中心的模型以及美国国家气象局的传统预报系统1。与前代产品相比，降雨预报准确度提升了60%1。
WeatherNext 3的核心改进体现在多个方面。预报分辨率从15-25平方公里升级至5平方公里1，相比谷歌前一代模型提高了五倍2。新模型能够每小时生成一次预报，而非行业标准的每6小时预报1。模型的参数量达到前代产品的2.4倍1。谷歌称其为首个直接整合原始观测数据的高分辨率全球预报AI模型1。
WeatherNext 3将被集成到谷歌搜索、谷歌地图和Gemini等产品中1，同时在Google云平台向用户和研究人员开放1。
Google DeepMind and Google Research have released WeatherNext 3, a new artificial intelligence weather prediction model that outperforms competing systems from Microsoft, NVIDIA, the European Centre for Medium-Range Weather Forecasts, and the U.S. National Weather Service. 1 The model achieved top accuracy in the Operational WeatherBench test, marking a significant advancement in AI-driven meteorological forecasting. 1
The upgraded model delivers substantial improvements in forecast precision and resolution. WeatherNext 3 increases forecast resolution to 5 square kilometers, up from the previous 15–25 square kilometer range, representing a fivefold improvement over its predecessor. 12 Rainfall prediction accuracy has improved by 60 percent compared to WeatherNext 2, and the model can now generate hourly forecasts rather than the standard six-hour intervals. 1 Notably, WeatherNext 3 is Google's first high-resolution global forecasting AI model to directly integrate raw observational data, allowing it to learn from real-time weather observations and extend predictions beyond the typical training data limitations of most global AI models. 12 The model's parameters have grown to 2.4 times the size of its predecessor. 1
Google plans to integrate WeatherNext 3 across its major products, including Google Search, Google Maps, and Gemini, while also making the model available on Google Cloud Platform for users and researchers. 1 According to Samier Merchant, a research engineer at Google Research, "This will be the first time some of these core variables are directly powering a large amount of Google products." 1 DeepMind research scientist Ferran Alet explained that machine learning addresses the challenge of "approximating noisy physics from incomplete information and limited computation." 1
评论
还没有评论，欢迎留下第一条。