Combining Remote Sensing and AI to Explore and Classify Volcanic Patterns

Abstract

Monitoring volcanic hazards is essential for understanding the behavior of active volcanoes, enhancing hazard forecasting, and mitigating potential impacts. The increasing availability of satellite sensors, particularly those providing thermal infrared data with varied spatial resolutions and revisit times, has significantly enhanced monitoring capabilities. However, the vast data volume requires advanced computational tools, such as Artificial Intelligence (AI) algorithms, to enable efficient, real-time processing and analysis. Recent AI advancements have shown considerable promise in enhancing satellite-based volcanic monitoring. Here, we introduce a cascading pipeline model designed to classify high-temperature volcanic features and quantify thermal anomalies using high-resolution Sentinel-2 Multi-Spectral Instrument (MSI) imagery. This automated approach allows for timely and detailed assessments of volcanic activity. We apply the model to three highly active volcanoes: Etna and Stromboli, in the Mediterranean, and Pacaya, in Central America. Etna and Stromboli, two of the most monitored volcanoes in the Mediterranean, offer complex eruption
patterns ideal for testing AI-based models. Pacaya’s frequent eruptions and lava flows provide valuable comparative data for testing the model’s robustness across different volcanic systems. Our objective is to classify volcanic spatial patterns and explore significant changes in thermal anomalies during periods of unrest, ultimately broadening the model’s applicability beyond the Mediterranean region.