The DeepMind study group has presented its new project that aims to help companies that need to predict rain. The tools we have now are more or less accurate for long periods of time, however it is not possible to predict whether or not it will rain within an hour or two.

These predictions are called “nowcast” and the DGMR (deep generative model of rainfall) tool aims to fill that gap with its artificial intelligence capable of recreating rainfall patterns.

The AI ​​takes this data and predicts.

Accurate predictions

The researchers’ paper was published today in Nature, there they explain how they have created an artificial intelligence that they “feed” with real data from the rain in real time. We talk about data such as the position of the clouds, their movement, the direction of the wind and more data that allows you to “learn” how some meteorological phenomena work.

After a while DGMR is able to reproduce what will happen next with quite certainty, specifically with the rain within the next two hours. Current tools can predict whether it will rain, but not intensity or location, while DGRM will be able to do both.

To test the accuracy of the DGRM data, the team asked 56 climatologists to choose which prediction method was most accurate. DGRM was compared with two other similar methods and 89% of the specialists preferred DGRM. The choice was made blind and none of the experts had participated in the project.

The team has been working on this project for several years, so they hope that its creation will be useful in the near future.