Deepmind AI convinces experts from the British betting service

Artificial intelligence cannot help us against bad weather. At least it can now predict extremely precisely when and where it will rain.

Science still finds it difficult to accurately predict rain or other weather phenomena within a short period of one to two hours. Now the Google sister company Deepmind wants to solve the problem with the help of AI. Your software should be able to accurately determine where and how heavily it should rain within the next 90 minutes. A first experimental test apparently ran convincingly.

Deepmind checked their weather forecast tool with experts from the British national weather service. That goes from one in the specialist journal Nature published study. According to the British meteorologists, the Deepmind software performed better than existing solutions in predicting the location, intensity and course of rain showers in almost 90 percent of all cases.

“While current practice focuses on quantitative improvements without worrying about operational benefits, we hope that this work will serve as the foundation for new data, coding and verification methodologies – as well as the deeper integration of machine learning and environmental science when predicting larger amounts of environmental variables, ”said the Deepmind researchers in their paper.

When and where it will rain depends on factors such as temperature, cloud formations and wind. Classic forecast models map all this data in the form of atmospheric simulations. Due to the many factors, these in turn require an enormous amount of computing effort, which is why they are particularly useful for longer-term predictions over several days.

However, in order to be able to make statements about the weather at short notice and without complex physical simulations, the Deepmind researchers have fed their AI with current radar data. The system was designed to generate new radar images from the imported radar images that are as similar as possible to the original data. Basically, there is a similar technique behind it, with which, for example, artificial portrait photos are also created. Only here no fake photos were generated, but radar images that predict the further course of the weather into the future.