Google DeepMind 发布了最新的人工智慧气象预测模型 WeatherNext 3。该模型不仅具备更优异的温度与降水预测能力，还新增了太阳辐射与云层覆盖估计，并且不再受限于政府气象机构每六小时更新一次的数据周期，能够每小时利用卫星图像持续刷新预报数据。

由于风能与太阳能正在重塑电网格局，天气状况日益直接决定能源供应，且电力交易市场已大幅转向以日内 15 至 30 分钟为单位的短周期交易。因此，WeatherNext 3 能够即时捕捉如云层遮蔽等局部天气变化，对能源交易商与公用事业机构预测再生能源发电量具有极高的应用价值。

传统 AI 气象模型多半依赖欧洲中期天气预报中心（ECMWF）每隔六小时产出且耗时五小时构建的数据集，导致预测具有时间滞后性。WeatherNext 3 则结合已有的 ECMWF 预报数据与即时三维大气卫星遥测影像，将典型气象预报中的数据时效延迟从约七小时大幅缩减至三到四小时。

Google DeepMind has unveiled its latest AI weather model, WeatherNext 3, which delivers improved temperature and precipitation predictions along with new solar and cloud cover estimates. By breaking free from the traditional six-hourly government data update cycles, the model refreshes forecasts every hour using incoming satellite imagery.

As wind and solar power continue to reshape global electricity grids, weather increasingly dictates energy supply, and power trading has shifted toward intraday 15- and 30-minute intervals. Consequently, WeatherNext 3's high-frequency updates are particularly valuable for energy traders and utility providers looking to forecast renewable power generation accurately.

Conventional AI weather models rely heavily on datasets from the European Centre for Medium-Range Weather Forecasts (ECMWF), which take hours to compile and update only every six hours. WeatherNext 3 integrates existing ECMWF forecast data with real-time three-dimensional atmospheric satellite images, reducing the information lag in forecasts from about seven hours to just three or four hours.