Satellites and Artificial Intelligence to monitor volcanoes: what is the Safari project of Ingv that also aims to make air traffic safer

The integration of the most advanced capabilities of satellite remote sensing and Artificial Intelligence (AI) is redefining volcano surveillance, opening up previously unexplored scenarios. The goal of this synergy is the fully automated characterization of the activity status of volcanic systems and, where possible, the prediction of their evolution.

At the heart of this scientific and technological transformation is SAFARI, an acronym for "An artificial intelligence-based StrAtegy For volcAno hazaRd monItoring from space". Launched in 2023 and coordinated by INGV in collaboration with national and international universities and research institutions, the project aims to develop an integrated strategy to assess volcanic hazard directly from space.

The strength of SAFARI lies in the fusion of an immense amount of information: satellite data acquired from optical, thermal, and microwave sensors are systematically compared with daily measurements from ground observation networks (seismic, geodetic, and geochemical).

This effort is supported by an international team composed of researchers from INGV, the University of Catania and Palermo, Max Planck Institute for Chemistry, Goma Volcanological Observatory, Instituto Geofísico in Ecuador, and Observatoire de Physique du Globe de Clermont-Ferrand.

The methodologies developed have been successfully applied to four key volcanoes: Etna and Vulcano in Italy, Nyiragongo in the Democratic Republic of the Congo, and Sangay in Ecuador. The latter two, with extremely high hazard and located in remote areas, rely heavily on space observation.

The results achieved in three years of activity address crucial junctions in volcanic monitoring. First of all, AI has been used to "clean" radar images. SAR (Synthetic Aperture Radar) data are often affected by noise and random fluctuations that generate granularity and complicate interpretation.

Thanks to an innovative filter, AI has "learned" to separate useful signals from noise, drastically improving the signal-to-noise ratio and reducing false anomalies. This technique, applied to the data from the Sentinel-1 satellite during the Sangay eruption in 2021, has produced a significant leap in the reliability of images.

A second milestone concerns the autonomous, real-time prediction of lava flow paths. The system automates flow propagation simulations to quickly generate hazard maps, which are essential for risk management. By using the estimated effusion rates from space, the model reproduces the extent of the front and anticipates the evolution of the lava field by days in advance, progressively improving the accuracy of risk maps.

Significant implications also emerge for air traffic safety. Knowing the exact height of a volcanic cloud is vital for aviation. On Etna, the SAFARI team trained a neural network on thousands of satellite images, resulting in a system capable of automatically distinguishing the colors of lava emissions from those of clouds generated by explosive activity and calculating the real-time height of the eruptive column.

Finally, in contexts with limited historical data, such as Nyiragongo, AI has operated with an “Unsupervised Learning” approach. By autonomously analyzing satellite data, the algorithm grouped the volcano's behaviors into “families” of activity, identifying the transition from calm to an anomalous phase right during the eruption in May 2021. In similar scenarios, AI can act as an “autonomous sentinel”, capable of quickly examining vast amounts of information and recognizing patterns imperceptible to the human eye. However, like any automated system, it always requires expert verification to prevent false alarms.

At the final meeting of the project, the results of this joint work were presented, marking a crucial transition: from the ideation and experimentation to the routine use of space technologies and artificial intelligence for the daily assessment of volcanic hazard on a global scale.