Machine learning algorithm that predicts lightning developed by German computer scientists

Lightning can have serious consequences not only for people and structures but also because they can upset the environment by triggering fires. Predicting thunderstorms with greater precision, even with regard to location, therefore remains of primary importance.

Jens Dittrich, professor of computer science at the University of Saarland, together with his student Christian Schön, therefore thought of developing software that could help in this regard. The two have thus developed a new algorithm that turns out to be more powerful than the previous ones and can predict thunderstorms with greater precision.

Beyond the precision level of this algorithm, this research is important because it explores the possibility of using artificial intelligence as regards the localized prediction of meteorological phenomena. And this is even more true for thunderstorms, a considerable precision is needed when they must be provided for in a specific region: the movement of cold and hot air masses must be detected in advance and with great precision.

The software is capable of using two-dimensional images, those produced by satellites, to detect movements of three-dimensional air masses. Lofa thanks to a new algorithm that basically calculates a future image. The algorithm has been trained with the machine learning technique to minimize errors. In the end, it turned out to be so precise that researchers are now able to calculate lightning and thunder with relative accuracy.

As the scientist reports, the algorithm, based only on satellite images, can predict lightning with a 96% accuracy in a forecast window that can last 15 minutes. In a five-hour forecast window, the degree of precision remains above 83%.

As Professor Dittrich explains, these are the results when the large masses of data that today’s tools, in this case the satellites, can provide us with today’s computational power are combined: computers are now able to recognize patterns that would remain entirely hidden from our eyes.