AI system can predict lightning strikes
Siobhan Treacy | November 08, 2019
Researchers from EPFL’s School of Engineering and Electromagnetic Compatibility Laboratory have created an artificial intelligence (AI) system that can predict lightning strikes. The system can predict lightning strikes within a 30 km (18.64 mile) radius to the nearest 10 to 30 minutes.
The system uses a combination of standard meteorological data and artificial intelligence to make its predictions.
Current systems are slow and complex, and require expensive external data from a radar or satellite. The new method uses data from weather stations. This allows the system to cover remote areas that are out of the range of radars and satellites. Data for the system is acquired easily and in real-time, which enables the system to send out weather alerts before lightning strikes.
The machine-learning algorithm was trained to recognize conditions that lead to lightning. To train the AI, researchers collected data from 12 Swiss weather stations from urban and mountainous environments for 10 years.
The algorithm took four parameters into consideration when making a prediction: atmospheric pressure, air temperature, relative humidity and wind speed. These parameters correlated with recordings from lightning detection and location systems. During training, the system learned the conditions through which lightning occurs.
The team tested its system and found success. After training, the system made correct predictions 80% of the time. This was the first time a system based on simple meteorological data has been used to predict lightning strikes with real-time calculations.
A paper on the technology was published in Climate and Atmosphere Science.
I can do that much more accurately using the XM satellite datalink and a $50 a month subscription, which I pay anyway. The downlink provides Nexrad at usually no more than a few minutes old. Combined with onboard radar, I can see exactly where weather conducive to lightning is presently at in relation to my projected flight path, and where I need to deviate in order to avoid it. The downlink also displays winds aloft, cell movement, cloud tops, lightning strikes, and echo tops. This can all be overlaid on top of the real time moving map. It allows me to form a mental picture of the meteorological conditions, including cell stages, movement, and other factors to predict in my head where not just the lightning will be, but hail and rain.
While I can see that this new AI might have beneficial applications in a number of areas, aviation is not one of them. Only a properly trained, competent, experienced pilot on board the aircraft is able to assure safe weather avoidance. I hope the research being conducted leads to a lot of learning and new discoveries, but from a pilot perspective, I can do a better job of predicting lightning than they can.