The model, called 'GenCast', outperforms traditional mid-range climate forecasts and is also able to better predict extreme weather conditions, tropical cyclone trajectories and wind energy production.

Details of the model were released in an article published in the magazine Nature, according to the agency Efe.

Having accurate weather forecasts is essential so that people, governments and organizations can make essential decisions in their daily lives, from carrying an umbrella to evaluating wind energy production or planning for extreme weather conditions to avoid disasters.

Traditional weather forecasts are based on numerical weather forecasting methods, which estimate the current weather and map it to a future weather forecast over time (known as deterministic forecasts), but this generates numerous potential scenarios, which are combined to produce a weather forecast.

Now, a team of scientists at Google has developed a machine learning weather forecasting method called GenCast that is capable of generating a probabilistic forecast, which predicts the likelihood of future weather based on current and past weather states.

The authors trained GenCast from 40 years (1979 to 2018) of data analysis of best estimates of climate occurrences.

Thanks to this training, the model is able to generate global forecasts for 15 days, for more than 80 atmospheric and surface variables, in eight minutes.

When compared to the European Center for Medium-Range Weather Forecasts (ENS) forecast suite - currently the best performing medium-range forecast worldwide - they found that GenCast outperformed ENS in 97.2% of 1,320 targets used.

GenCast is also more effective at predicting extreme weather conditions, tropical cyclone trajectories, and wind energy production.

The authors argue that GenCast can generate more efficient and effective weather forecasts to support effective planning.