由 Google DeepMind 和 Google Research 开发的人工智慧模型 WeatherNext，在飓风梅丽莎登陆牙买加前五天，就以高达 80% 的置信度准确预测了它将发展为五级飓风。这项在《自然》杂志上发表的研究指出，该模型能以史无前例的准确度预测气旋，平均为预报员争取到比现有模型多一天的预警时间。这额外的一天对于安排疏散和调度资源等决策至关重要，能有效帮助减轻灾情。

预测极端天气对人工智慧来说是一大挑战，因为机器学习通常需要大量数据，而极端天气相对罕见。此外，飓风预测不仅需要全球尺度的数据来预测路径，还需要局部海洋与大气条件的数据来预测强度。然而，WeatherNext 即使仅使用较低解析度的输入数据，依然能在路径和强度的预测上展现出卓越的表现，甚至让预报中心首次能在风暴仅为一级阶段时，就成功预测其将快速增强为五级飓风。（关键数字：5, 1）

除了提供单一预测，WeatherNext 还能为发展中的风暴生成多达 1000 种潜在情境，帮助捕捉微小偏差可能带来的「蝴蝶效应」，这是依赖现有计算能力的传统数值模型难以实现的。尽管 AI 模型带来重大突破，专家仍强调人类在将这些预测转化为具体影响评估上的关键角色。目前 Google DeepMind 已开源该模型，期望借由开放给研究社群使用，能进一步揭开气旋运作的未知科学机制。

Developed by Google DeepMind and Google Research, the artificial intelligence model WeatherNext accurately predicted that Hurricane Melissa would hit Jamaica as a Category 5 hurricane five days before landfall with 80 percent confidence. According to a study published in Nature, the model predicts cyclones with unprecedented accuracy, giving forecasters an average of one more day of lead time compared to existing models. This crucial extra day is extremely valuable for making time-sensitive decisions like organizing evacuations and deploying resources, ultimately helping to mitigate disasters.

Forecasting extreme weather events poses a significant challenge for AI since machine learning requires ample data, and such events are rare. Furthermore, predicting hurricanes requires multi-scale data: global data for the storm's track and local atmospheric and oceanic data for its intensity. However, WeatherNext excels at predicting both track and intensity despite using relatively coarse-resolution input data. Its outstanding performance even allowed forecasters to predict a Category 5 hurricane when the storm was only at a Category 1 stage for the first time.

Instead of providing a single forecast, WeatherNext generates up to 1,000 potential scenarios for a developing storm, helping forecasters account for the butterfly effect where small deviations lead to significant changes—a feat impossible with current numerical models and computing power. While the AI model represents a major breakthrough, experts emphasize that human interpretation remains critical in translating these forecasts into actual real-world impacts. Google DeepMind has now open-sourced the model, hoping that giving the research community access will help uncover new scientific insights into how cyclones operate.