A new study from University of Washington (UW) has shown that machine learning can be used to improve forecasts for lightning. Lightning is a destructive force that has the potential to cause extensive damage to infrastructure, buildings and even create huge fires such as the massive California Lightning Complex fires. Having the ability to prepare for potential lightning forecasts could lead to better readiness for wildfires, improve warning times and create longer climate models. "The best subjects for machine learning are things that we don't fully understand," said Daehyun Kim, an associate professor of atmospheric sciences at UW. "And what is something in the atmospheric sciences field that remains poorly understood? To our knowledge, our work is the first to demonstrate that machine learning algorithms can work for lightning."

Lightning is one of the most destructive forces of nature, as in 2020 when it sparked the massive California Lightning Complex fires, but it remains hard to predict. A new study led by the University of Washington shows that machine learning – computer algorithms that improve themselves without direct programming by humans – can be used to improve lightning forecasts. Better lightning forecasts could help to prepare for potential wildfires, improve safety warnings for lightning and create more accurate long-range climate models. "The best subjects for machine learning are things that we don't fully understand. And what is something in the atmospheric sciences field that remains poorly understood? Lightning," said Daehyun Kim, a UW associate professor of atmospheric sciences.

Lightning is one of the most destructive forces of nature, as in 2020 when it sparked the massive California Lightning Complex fires, but it remains hard to predict. A new study led by the University of Washington shows that machine learning -- computer algorithms that improve themselves without direct programming by humans -- can be used to improve lightning forecasts. Better lightning forecasts could help to prepare for potential wildfires, improve safety warnings for lightning and create more accurate long-range climate models. "The best subjects for machine learning are things that we don't fully understand. And what is something in the atmospheric sciences field that remains poorly understood? Lightning," said Daehyun Kim, a UW associate professor of atmospheric sciences.

Lightning is one of the most destructive forces of nature, as in 2020 when it sparked the massive California Lightning Complex fires, but it remains hard to predict. A new study led by the University of Washington shows that machine learning--computer algorithms that improve themselves without direct programming by humans--can be used to improve lightning forecasts. Better lightning forecasts could help to prepare for potential wildfires, improve safety warnings for lightning and create more accurate long-range climate models. "The best subjects for machine learning are things that we don't fully understand. And what is something in the atmospheric sciences field that remains poorly understood? Lightning," said Daehyun Kim, a UW associate professor of atmospheric sciences.

As one of the most irregular phenomena in nature, lightning is very disturbing. Scientists have lately made an AI system that forecasts lightning up to 30 minutes before it strikes. Lightning regularly kills animals and people, initiates fires, destroys power lines and keeps aircraft stranded. Till now, it has been almost out of the question to predict lightning, with no simple technology for predicting where and when it will strike the earth. Engineers at the Ecole Polytechnique Federale de Lausanne's (EPFL) School of Engineering built a simple and cheap system to forecast when lightning will strike.

Despite the progress within the last decades, weather forecasting is still a challenging and computationally expensive task. Current satellite-based approaches to predict thunderstorms are usually based on the analysis of the observed brightness temperatures in different spectral channels and emit a warning if a critical threshold is reached. Recent progress in data science however demonstrates that machine learning can be successfully applied to many research fields in science, especially in areas dealing with large datasets. We therefore present a new approach to the problem of predicting thunderstorms based on machine learning. The core idea of our work is to use the error of two-dimensional optical flow algorithms applied to images of meteorological satellites as a feature for machine learning models. We interpret that optical flow error as an indication of convection potentially leading to thunderstorms and lightning. To factor in spatial proximity we use various manual convolution steps. We also consider effects such as the time of day or the geographic location. We train different tree classifier models as well as a neural network to predict lightning within the next few hours (called nowcasting in meteorology) based on these features. In our evaluation section we compare the predictive power of the different models and the impact of different features on the classification result. Our results show a high accuracy of 96% for predictions over the next 15 minutes which slightly decreases with increasing forecast period but still remains above 83% for forecasts of up to five hours. The high false positive rate of nearly 6% however needs further investigation to allow for an operational use of our approach.