Global medium-range weather forecasting is critical to decision-making across many social and economic domains. Traditional numerical weather prediction uses increased compute resources to improve forecast accuracy, but cannot directly use historical weather data to improve the underlying model. We introduce a machine learning-based method called "GraphCast", which can be trained directly from reanalysis data. It predicts hundreds of weather variables, over 10 days at 0.25 degree resolution globally, in under one minute. We show that GraphCast significantly outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets, and its forecasts support better severe event prediction, including tropical cyclones, atmospheric rivers, and extreme temperatures. GraphCast is a key advance in accurate and efficient weather forecasting, and helps realize the promise of machine learning for modeling complex dynamical systems.
核心贡献 · Key contributions
提出 GraphCast，一种直接从再分析数据训练的全球中期天气预报机器学习方法。 Introduces GraphCast, a machine learning method for global medium-range weather forecasting trained directly from reanalysis data.
在单个 TPU v4 设备上不到一分钟内实现全球 0.25°分辨率 10 天预报。 Achieves 10-day forecasts at 0.25° resolution globally in under one minute on a single TPU v4 device.
在 1380 个验证目标中，90%优于最准确的操作确定性系统（HRES）。 Outperforms the most accurate operational deterministic system (HRES) on 90% of 1380 verification targets.
展示了对热带气旋、大气河流和极端温度等极端事件预测的改进。 Demonstrates improved prediction of severe events including tropical cyclones, atmospheric rivers, and extreme temperatures.
采用编码-处理-解码图神经网络架构，共 3670 万个参数。 Uses an encode-process-decode graph neural network architecture with 36.7 million parameters.
表明机器学习天气预报可以补充传统数值天气预报方法。 Shows that machine learning weather prediction can complement traditional numerical weather prediction methods.
局限 · Limitations
由于训练数据和硬件限制，以 0.25°分辨率和 6 小时时间步长运行。 Operates at 0.25° resolution and 6-hour time steps due to training data and hardware constraints.
专注于确定性预报，不如集合方法能捕捉不确定性。 Focuses on deterministic forecasts, not capturing uncertainty as well as ensemble methods.
MSE 训练目标在较长预报时效鼓励空间模糊，限制了某些应用的价值。 MSE training objective encourages spatial blurring at longer lead times, limiting value for some applications.
关键依赖于来自数值天气预报系统的高质量再分析数据。 Depends critically on high-quality reanalysis data from numerical weather prediction systems.
不能替代传统方法；而是补充传统方法，并具有进一步规模扩张的潜力。 Not a replacement for traditional methods; complements them with potential for further scaling.