rotative
wired ai · 2026年8月6日

DeepMind 稱其開源天氣模型在颶風預報上提供關鍵 24 小時優勢

谷歌 deepmind 的 'weathernext' 模型比傳統方法提前整整一天預測氣旋,並將開源。

read in English · 简体中文 · 日本語 · 한국어

deepmind's weathernext model, developed in collaboration with google research, has demonstrated the ability to predict cyclone intensity and track with accuracy that effectively grants forecasters an extra 24 hours of warning compared to previous models. the model successfully predicted the catastrophic hurricane melissa in october 2025, identifying it as a category 5 threat five days before landfall when it was only a category 1 storm. unlike previous ai models that struggled with intensity due to a lack of specific data, weathernext was trained on global weather patterns to infer local cyclone behavior. the model will be open-sourced, potentially revolutionizing disaster preparedness despite researchers admitting they do not yet fully understand the internal mechanisms of its predictions.

read at source ↗

← rotative