For nearly as long as the modern computer has existed, it has been used to forecast the weather. First used during World War II to simulate (模拟) nuclear weapons, computers were soon adopted to simulate the future state of the atmosphere, creating the modern discipline of weather prediction. Although that discipline has grown ever more complicated and now produces reliable forecasts several weeks in advance, its approach remains the same: using large amounts of calculating power to solve equations (方程).
Over the past year, artificial intelligence (AI) has begun to change that. Tech companies including Google and Nvidia have trained AI models to predict the weather up to 10 days in advance, with an accuracy equaling or even topping traditional models — and with far less calculation overhead. Rather than solving equations, these AI models predict the near future based on patterns learned through training on 40 years of past weather, which is recorded by the traditional model of the European Centre for Medium-Range Weather Forecasts (ECMWF), the world’s top weather agency. Once trained, the AI models can work out a forecast on a computer in 1 minute rather than taking 2 hours to run on a supercomputer.
ECMWF has already begun to produce its own AI forecast, and other weather agencies are eager to catch up. The new models aren’t perfect. They struggle to predict certain essential features—hurricane intensity, for example. But AI forecasters will only improve as they begin to learn from direct weather observations collected by sensors, not just data already passed through existing models. Besides, their speed could allow agencies to run them over and over, as they capture in the atmosphere the full spread of uncertainty, be it necessary or unnecessary for weather prediction.
No one expects traditional weather prediction to disappear. Another branch, climate models, an extension of weather forecasting, for example, rely on equation solving just as traditional weather models do. But in the long term, the output of climate models may itself become training data for a climate forecasting AI, which might ultimately do a better job than the traditional models.
【小题1】How do AI models predict weather?| A．By running on a supercomputer. | B．By recording traditional models. |
| C．By working on the existing data. | D．By making massive calculations. |
| A．They may be overly operated. | B．They may be slow to respond. |
| C．They may confuse natural disasters. | D．They may bring unfair competitions. |
| A．They lack accurate data. | B．They need intensive training. |
| C．They work in a traditional way. | D．They determine weather forecasting. |
| A．How Can AI Aid Atmosphere Study? |
| B．Should We Trust AI to Predict Hurricanes? |
| C．Weather Forecast Is Having an AI Moment |
| D．Tech Giants Are Competing in Data Collection |
