The weather is at the centre of many of our actions and conversations. Inaccurate forecasts can thus stir up our planning, all while giving us something to complain about, of course. But thanks to Google’s DeepMind artificial intelligence research lab, they might be something of the past.
DeepMind has presented GenCast in a study published in the journal Nature. The AI-based weather forecast model is said to make faster and more accurate 15-day weather predictions than anything we’ve seen before and it has a unique method to do so.
“GenCast is a machine learning-based weather model, which learns directly from historical weather data. This is in contrast to traditional models, which make forecasts by solving physics equations,” Ilan Price, DeepMind researcher and a co-author of the study, told the Register’s Thomas Claburn. “One limitation of these traditional models is that the equations they solve are only approximations of the atmospheric dynamics. GenCast is not limited to learning dynamics/patterns that are known exactly and can be written down in an equation.”
More precisely, GenCast bases its predictions on historical weather data ranging from 1979 to 2018 from the European Centre for Medium-Range Weather Forecasts (ECMWF). The data comprises multiple sets of information, including temperature, but also wind speed and air pressure measurements at different altitudes. In order to test the AI forecast, GenCast was asked to make 15-day forecast predictions on 1,320 weather events from 2019. Those were then compared to predictions by the ECMWF and the actual measurements. Overall, GenCast was more accurate than the ECMWF in 97.2% of the cases.
However, GenCast isn’t live yet and probably won’t be for a little longer. And there’s a good reason for that. As AI-based weather forecast models base their predictions on past events, with the current effects of climate change, it is hard for them to predict extreme weather, although improvements are being made. Tropical cyclones and especially their intensity, however, remain a problem point. In an interview with CNN, Ilan Price, the lead author of the study and a senior research scientist with DeepMind, explained the issue should nevertheless be solved in the near future.
“You can be as skeptical as you want against machine learning forecasts in principle”, Peter Dueben, a machine learning expert and head of Earth system modeling at the ECMWF, told CNN. “These models will make a positive impact on our weather predictions; there’s no question there.”
