内容来源：https://aibusiness.com/foundation-models/nvidia-ai-weather-forecast-models

内容总结：

英伟达近日宣布，其人工智能气象预报平台Earth-2迎来重要升级，推出三款全新开源模型，旨在通过AI技术显著提升气象预测的精度与效率。

此次发布的新模型包括：专注于15天中期高精度预报的Earth-2 Medium Range，可覆盖温度、气压、风场等70余种气象变量；基于生成式AI模型StormScope的Earth-2 Nowcasting，能提供公里级分辨率的局地强天气短临预报；以及采用HealDA模型的Earth-2 Global Data Assimilation，可在数秒内生成全球大气初始场，为预报提供关键数据基础。

这些模型将与平台现有开源模型CorrDiff、FourCastNet3协同工作，构成全球首个完全开源、加速化的AI气象软件栈。英伟达表示，该平台通过预训练模型和开源库，能将从数据同化到两周全球预报的全流程提速，大幅降低传统超算物理模型的高昂计算成本。

目前，以色列气象局、台湾中央气象局、美国国家气象局、天气公司以及道达尔能源、埃尼集团、安盛保险、标普全球能源等企业和机构已开始应用相关技术。道达尔能源气象产品负责人表示，Earth-2 Nowcasting等模型能提升能源系统短期风险感知能力，对业务决策具有突破性意义。

全套模型将于发布后在GitHub和Hugging Face平台开源，供商业及非商业用途使用。

中文翻译：

由谷歌云赞助
如何选择首个生成式AI应用场景
要启动生成式AI项目，首先应关注能够优化人类信息交互体验的领域。

这些模型是英伟达Earth-2平台的核心组件，该平台被定位为全球首个完全开源的气象AI软件栈。本周，英伟达为其Earth-2系列AI模型发布了新功能，称其为全球首个完全开源、加速运算的气象AI软件栈。

该平台整合了多种模型、程序库与框架。英伟达表示，这将使气象与气候AI技术比以往更易普及，潜在用户涵盖科学家、初创企业、开发者乃至政府机构。通过调用这些预训练模型与程序库，用户可加速从初始观测数据处理到生成两周全球预报的完整预测流程。

英伟达在发布博客中写道："传统天气预报依赖运行物理模型的大型超级计算机。AI驱动的天气预报能显著节约计算时间与成本，让更多国家、气象机构和企业能够运行特定场景的预报系统。"

具体而言，英伟达宣布在工具集中新增三款模型：Earth-2中期预报模型、Earth-2临近预报模型及Earth-2全球数据同化模型。

Earth-2中期预报模型基于英伟达全新Atlas架构构建，可提供15天高精度天气预报。该模型专为中期预测设计，涵盖温度、气压、风速、湿度等70余项气象变量。

Earth-2临近预报模型采用名为StormScope的生成式AI技术，可为局部风暴及灾害性天气提供公里级精度的预报。

Earth-2全球数据同化模型由HealDA模型驱动，能生成天气预报所需的大气初始状态。该系统使用GPU替代传统超算，可在数秒内生成全球实时气象快照。英伟达指出，与Earth-2系列模型结合后，这一变革将同时提升预报速度与精度。

这些新模型将与英伟达现有开源气象模型CorrDiff和FourCastNet3协同工作——后两者分别用于提升预报分辨率与预测精度。

早期用户包括AI气象初创企业Brightband，以及以色列气象局、台湾中央气象局、天气频道公司、美国国家气象局等国家级与商业气象机构。道达尔能源、埃尼集团、协鑫集团、西南电力联营体（与日立合作），以及安盛保险、标普全球能源资讯等金融风险与情报机构也已开展早期应用。

道达尔能源气候与天气预报产品经理埃马纽埃尔·勒博尔涅在博客中表示："英伟达Earth-2标志着先进气象智能大规模应用迈出关键一步。Earth-2临近预报等模型对我们的业务具有突破性意义，它们能提升能源系统对分钟级变化与局部影响的短期风险感知与决策能力。"

英伟达表示，完整模型系列将开放商业与非商业授权，发布后可通过GitHub和Hugging Face平台获取。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The models come as part of the company's Earth-2 platform, pitched as the first fully open AI weather software stack.
Nvidia has this week unveiled new capabilities for Earth-2, a family of AI models that the company described as the world's first fully open, accelerated weather AI software stack.
The platform is a collection of models, libraries and frameworks that Nvidia said make weather and climate AI more accessible than before, with potential users spanning scientists, startups and developers to enterprises and governments.
With access to these pretrained models and libraries, Nvidia said users can accelerate every step of forecasting, from processing initial observation data to generating two-week global forecasts.
"Historically, weather forecasting has relied on powerful supercomputers running physics-based models," Nvidia wrote in a blog post about the launch. "AI-powered weather forecasting saves significant computational time and costs, allowing more nations, weather enterprises and businesses to run application-specific forecasting systems."
Specifically, Nvidia has announced three new models' availability in the toolset; Earth-2 Medium Range, Earth-2 Nowcasting and Earth-2 Global Data Assimilation.
Earth-2 Medium Range is a 15-day, high-accuracy forecasting model built on Nvidia's new Atlas architecture. Designed for medium-range prediction, the model covers more than 70 weather variables, including temperature, pressure, wind and humidity.
Earth-2 Nowcasting uses a generative AI model called StormScope to deliver kilometer-scale forecasts for local storms and hazardous weather.
Earth-2 Global Data Assimilation, powered by the HealDA model, generates the initial atmospheric conditions required for weather prediction. Using GPUs rather than traditional supercomputers, the system can produce global snapshots of current conditions in seconds, a shift Nvidia says improves both forecasting speed and accuracy when paired with Earth-2 models.
These join Nvidia's existing Earth-2 open weather models CorrDiff and FourCastNet3, designed for added reporting resolution and forecasting accuracy, respectively.
Early customers include AI weather startup Brightband, along with national and commercial forecasters such as the Israel Meteorological Service, Taiwan's Central Weather Administration, The Weather Company and the U.S. National Weather Service.
TotalEnergies, Eni, GCL and Southwest Powerpool in collaboration with Hitachi, as well as financial risk and intelligence firms AXA and S&P Global Energy are also early use cases.
"Nvidia Earth-2 represents a major step forward in how advanced weather intelligence can be operationalized at scale," said Emmanuel Le Borgne, climate and weather forecast product manager at TotalEnergies SE, in the blog post. "Models like Earth-2 Nowcasting are groundbreaking for our business because they improve short-term risk awareness and decision-making in energy systems where minutes and local impacts matter."
Nvidia said the full model family will be licensed for commercial and non-commercial use and will be available on GitHub and Hugging Face upon release.