Scientists say they've developed a system using machine learning to predict when and where lighting will strike. The research was led by engineers from the Swiss Federal Institute of Technology in Lausanne, Switzerland.
European researchers have estimated that between 6,000 and 24,000 people are killed by lightning worldwide each year. The strikes can also cause power outages, destroy property, damage electrical equipment and start forest fires. For these reasons, climate scientists have long sought to develop methods to predict and control lightning. In the United States and other places, ground-based sensing devices are used to identify strikes as they happen. But, no system has been created to effectively predict lightning.
The system tested in the experiments used a combination of data from weather stations and machine learning methods. The researchers developed a prediction model that was trained to recognize weather conditions that were likely to cause lightning.
The model was created with data collected over a 12-year period from 12 Swiss weather stations in cities and mountain areas. The data related to four main surface conditions: air pressure, air temperature, relative humidity and wind speed.
The atmospheric, data was placed into a machine learning algorithm (算法), which compared it to records of lightning strikes. Researchers say the algorithm was then able to learn the conditions under which lightning happens.
Amirhossein Mostajabi is a PhD student at the institute who led the development of the method. He said, "current systems for gathering such data are slow and complex and require costly collection equipment like radar or satellites. "
"Our method uses data that can be obtained from any weather station," Mostajabi said. "This will improve data collection in very remote areas not covered by radar and satellite or in places where communication systems have been cut," he added.
