Name: AI 打敗傳統氣象模型？NVIDIA Earth-2 重新定義天氣預測！ (NVIDIA Earth-2: The Future of AI Weather Forecasting Is Open)
Uploaded: 2026-02-07T21:01:20.000Z
Duration: 2 min 20 s

Weather has a profound impact on our daily lives.

天氣對我們的日常生活有著深遠影響。

Predicting how it will affect agriculture, energy,

預測它將如何影響農業、能源，

aviation, and emergency response is critical.

航空與緊急應變系統，至關重要。

But the atmosphere is chaotic, and forecasting is complex.

然而，大氣系統本身充滿混沌，天氣預測極為複雜。

It accelerates the AI revolution in forecasting

它加速了天氣預測領域的 AI 革命，

by offering developers and researchers an open set of tools that lowers the barrier to AI-enabled weather simulation.

為開發者與研究人員提供一套開放工具，大幅降低 AI 天氣模擬的門檻。

Building on the impact of models like ForeCastNet and CoreDiff,

在 ForeCastNet 與 CoreDiff 等模型成果的基礎上，

NVIDIA Earth-2 is now a complete portfolio of open models.

NVIDIA Earth-2 如今已成為完整的開放模型組合。

Introducing Earth-2 Global Data Assimilation,

包含 Earth-2 全球資料同化模型，

To initialize predictions, Earth-2 Global Data Assimilation models turn observations from diverse sources

為了啟動預測，Earth-2 全球資料同化模型可將來自多元來源的觀測資料，

— like balloons, satellites, and stations — into atmospheric states in seconds,

如氣球、衛星與地面測站，在數秒內轉換為大氣狀態，

a task that previously took hours on high-performance computing clusters.

這項工作過去需要在高效能運算叢集上耗費數小時。

Earth-2 Nowcasting models offer severe weather prediction at continental scale from zero to six hours.

Earth-2 即時預報模型可在 0 到 6 小時內，提供洲際尺度的劇烈天氣預測。

They represent the first time AI has surpassed the accuracy of physics-based forecasting of storm evolution.

這是 AI 首次在風暴演變預測準確度上，超越傳統物理模型。

Earth-2 Medium Range models deliver 15-day global forecasts with accuracy that outperforms other open models on standard benchmarks across more than 70 variables.

Earth-2 中期預報模型可提供長達 15 天的全球天氣預測， 在超過 70 項變數的標準測試中，其準確度勝過其他開放模型。

Earth-2’s models span the complete forecasting pipeline.

Earth-2 的模型涵蓋完整的天氣預測流程。

When they’re combined, weather can be simulated at speeds previously impossible,

當這些模型結合使用時，天氣模擬的速度將達到前所未有的程度，

以分鐘取代過去需要數小時甚至數天的運算。

And because Earth-2 is completely open — not just model weights, but inference tools and training frameworks too —

而且，Earth-2 完全開放——不只模型權重， 也包含推論工具與訓練框架——

developers can customize and fine-tune simulations to specific needs,

讓開發者能依照特定需求，自由客製與微調模擬內容，

using their own data and infrastructure to build sovereign weather and climate predictions with full control.

使用自有資料與基礎架構，打造具備完整主控權的自主天氣與氣候預測系統。

See into the future of weather with NVIDIA Earth-2.

透過 NVIDIA Earth-2，看見天氣的未來。