Knowing the weather is essential for making multiple daily decisions, but as the weather becomes more extreme, as the DANA disaster in Valencia has shown, it also influences our safety. Until now, accurate and reliable forecasts were limited to one week, but a new Artificial Intelligence (AI) from the company DeepMind of Google in London seems to have overcome that barrier.
GenCast is capable of quickly and reliably forecasting the weather up to fifteen days in advance, as the company announced in the magazine ‘Nature’ this week. This exceeds the traditional predictions carried out by the IFS model of the European Center for Medium-Range Weather Forecasts (ECMWF), considered the best in the world. In addition, Google’s AI shines in predicting extreme weather and the trajectories of tropical cyclones, which could save the lives of thousands of people, and is capable of anticipating what wind energy production will look like.
Traditional weather forecasts are deterministic, they are based on numerical methods, which estimate the current weather and relate it to a future forecast. The new machine learning model, however, is probabilistic: it predicts the probability of future weather conditions based on current and previous states. The authors, led by researcher Rémi Lam, trained GenCast with 40 years (from 1979 to 2018) of analysis data of the best estimates of weather phenomena, being able to generate 15-day global forecasts, in steps of 12 hours, for more than 80 atmospheric and surface variables, in 8 minutes. When compared to the work of ECMWF, a service that 35 countries rely on to generate their own forecasts, they found that GenCast outperformed it on 97.2% of the 1,320 targets used, such as wind speeds, temperatures and other variables.
GenCast is significantly ahead of DeepMind’s previous artificial intelligence weather program, introduced in 2023 and which advanced the time by ten days ahead.
DeepMind’s achievement comes two months after the company’s other artificial intelligence researchers won the Nobel Prize in Chemistry for predicting the structure of proteins, providing two examples of how machine learning, which raises many fears, It can be a useful tool for science.
For those responsible for the study, GenCast’s more accurate forecasts “can help officials protect more lives, prevent harm and save money.” Getting “better and more advanced warnings about where a hurricane will make landfall is invaluable.”
#Google #surpasses #weather #predictions #fifteen #days #ahead
