Scientists at the University of Lausanne have created artificial intelligence (AI) that can predict lightning strikes 30 seconds before this event only on the basis of weather data. Description of the development published in the journal Climate and Atmospheric Science.

The neural network was trained on the data on in what types of clouds and under what meteorological conditions similar discharges occur according to the gradient boosting technique. The technique is based on a system of relatively inaccurate predictive neural networks that learn from the mistakes of their predecessors.

Then, data from several tens of weather stations were loaded into the neural network - on their basis, the AI ​​learned to predict lightning from characteristic changes in temperature, humidity and air pressure, as well as other weather data. Forecast accuracy is still low and amounts to 76%.

Earlier, researchers from the United States used machine learning to understand the physics of earthquakes and identify signs of impending natural phenomena. After successfully predicting laboratory earthquakes, a group of geophysicists applied a machine learning algorithm in the Northwest Pacific.

hightech.fm