Over the warm waters of the Indian Ocean, every one and a half to two months, a huge wave of moist air gathers. It moves eastward at a pedestrian's pace and takes forty to ninety days to circle half the planet. This wave influences when rains begin in India and whether frosts strike Europe. But the wave is capricious: it sometimes strengthens, sometimes disappears, and is hard to predict.
Scientists have devised a way to understand this capriciousness. They reduced the wave to a few numbers, much as a puppeteer holds just a few strings. One number tells the wave's phase, another how strong it is. Then they trained a neural network to complete the full picture of clouds and winds from these numbers. The network learned from a sixty-year weather record gathered from satellites and weather stations. Now you only need to input the numbers, and within half an hour it will draw sixty years of plausible tropical waves.
This kind of puppet theater is the first to link simple numbers to the full complexity of the atmosphere. Previously, scientists either calculated only the numbers and lost cloud details, or ran huge models and drowned in minutiae. Now the neural network serves as a bridge: it takes simple commands and returns a rich picture. This helps better predict tropical weather and test why the wave behaves as it does.
Generating a 60-year MJO record takes about 30 minutes, comparable in computational cost to models of intermediate complexity. Moreover, the model can create counterfactual scenarios, such as a perpetual MJO with a constant period, which are impossible to observe in nature.
The idea of 'prompting' a climate model resembles weather control in stories where scientists set parameters and obtain predictable yet realistic responses. However, here there is no manipulation of the real atmosphere, only exploration of its virtual twins.