Progress in weather analysis opens the way to tangible benefits for society, including protecting lives and infrastructure through extreme event monitoring and early warning services. By training deep learning models to detect storm patterns in satellite images, the team of CMCC scientists has harnessed the power of artificial intelligence and satellite technology. The deep learning approach, combined with 24/7 satellite monitoring, enables the identification of where lightning is likely to occur, boosting the effectiveness and timeliness of weather alerts. “This research opens a new direction for severe-weather monitoring and nowcasting,” the scientists say, commenting on the research, which is an outcome of an initiative of the Italian government, in collaboration with the European Space Agency, and the full exploitation of CMCC’s high-performance computing infrastructure, which provided the computational resources needed to train the deep-learning model on a large dataset.
Lightning poses a direct threat to human life and can cause serious damage to critical infrastructure – from power grids and telecommunications to aviation operations – while also triggering wildfires, a particularly relevant risk in Mediterranean environments. Being able to identify, almost in real time, the areas where lightning is occurring and severe storms are underway is therefore closely linked to safety.
A new study by CMCC researchers uses a deep learning approach to identify where lightning is likely occurring over Italy. The approach can support early-warning systems for outdoor activities and for organizations such as the Civil Protection Agency, as well as for aviation and energy sectors.
“Being able to identify in near real time the areas where lightning is occurring and severe convective phenomena are underway helps improve our understanding of these events, an essential first step toward nowcasting – the real-time, highly detailed observation and short-term forecasting of weather conditions,” explains CMCC researcher and team coordinator of the study Paola Mercogliano.
The core innovation is a fully satellite-based lightning detection method using deep learning and 24/7 monitoring using satellites. This offers spatially consistent coverage across the entire Italian territory, including areas where ground-based sensor networks are sparse, such as mountainous regions. By focusing on three specific infrared channel differences known to highlight deep convection signatures, the CMCC team designed an AI model that learns to recognize patterns associated with electrically active storms directly from satellite images.
“This research demonstrates what is possible when AI is combined with satellite data, opening a new direction for severe-weather monitoring and nowcasting,” says lead author of the study and CMCC researcher Paolo Duminuco. “Thanks to their extensive coverage and temporal consistency, satellite observations are a powerful complement to ground-based networks. This work can also help motivate more researchers to explore how satellite data can be fully exploited through advanced AI methods.”
A key methodological challenge in detecting lightning from satellite data is that lightning is rare – only about 0.6% of the samples in their dataset correspond to lightning events. The authors addressed this by designing a training strategy that gives more weight to rare events, encouraging the model to recognize and correctly classify the minority of samples where lightning occurs while keeping false alarms under control.
Over a one-year test period, the model demonstrates strong overall skill, producing probability maps that highlight where lightning is likely occurring and intercepting a large fraction of lightning events with relatively limited false alarms.
A key enabling factor was CMCC’s high-performance computing infrastructure, which provided the computational resources needed to train the deep-learning model on a large, high-frequency SEVIRI dataset. The activity was carried out within the IRIDE program, a national initiative launched by the Italian government in collaboration with the European Space Agency (ESA).
“This activity is an early milestone in CMCC’s strategic effort to strengthen nowcasting capabilities by combining AI with satellite observations, complementing ground-based monitoring, with the long-term goal of developing extreme-weather nowcasting methods that can evolve into real-time services for multiple sectors,” says Mercogliano.
By leveraging AI and satellite observations together, CMCC is contributing to building the next generation of tools to better understand and manage weather and climate risks in Italy and beyond. The study positions CMCC at the cutting edge of research at the intersection of AI, satellite meteorology, and climate-related risk, with a methodology that is technically advanced, operationally oriented, and strategically aligned with CMCC’s mission to provide science-based information and services that help reduce the impacts of extreme events.
For more information:
F. Duminuco et al., “A Convolutional Neural Network for Lightning Strikes Detection Over the Italian Territory Using SEVIRI@MSG Data,” in IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1-13, 2025, Art no. 4114513, doi: 10.1109/TGRS.2025.3641256.
