AI weatherman: more accurate, faster, more opaque

A snapshot consisting of countless measurements of temperature, wind speed, air pressure and humidity. In addition, a fine grid spanning an entire continent in three dimensions.

From such a gigantic database, a supercomputer with access to millions of computing cores calculates the weather forecast for the next few days using well-researched equation systems.

This requires an enormous amount of electricity and takes several hours despite the impressive computing power. And in the end, there is only one weather forecast, which is very accurate, but not necessarily correct. It should be noted, however, that major trends like heatwaves or storms can be forecast more reliably than the local weather at a specific location.

The European Centre for Medium-Range Weather Forecasts (ECMWF) has now introduced an exciting shortcut to obtain a comparably good outlook on our weather.

For this purpose, an AI was trained with European weather data from the last 40 years. The result is a comparatively lean bundle of algorithms that calculate astonishingly precise forecasts.

According to those involved at the ECMWF, the reliability of the forecasts is on a par with the weather models currently in use. However, so little computing capacity is required, you can create them on a simple PC for home use.

Climate change complicates things again

It took four weeks of AI training to produce this impressively fast model. The result is fascinating, even if the researchers themselves admit that they are not yet able to fully understand how their artificial intelligence works when it comes to weather forecasting. Of course, this does not detract from the success.

At best, this will result in faster forecasts that are also more accurate for more limited regions and do not require gigantic computing capacities.

However, the system will probably not be able to replace the work of meteorologists for the time being. Because the weather has changed significantly over the last 40 years (keyword: climate change), certain trends and feedback effects must also be taken into account in order to keep the forecasts reliable.

In any case, AI can provide support because it produces results a hundred times faster and thus helps to create long-term forecasts for the future, which would otherwise take far too long to calculate. The only thing that remains to be understood is why the calculation is so fast.