Monitoring Earth's atmosphere with Sentinel-5 TROPOMI and Artificial Intelligence: Quantifying volcanic SO2 emissions

Identifying changes in volcanic unrest and tracking eruptive activity are fundamental for volcanic surveillance and monitoring. Magmatic gases, particularly sulphur dioxide (SO₂), play a crucial role in influencing eruptive styles, making the monitoring of SO₂ emissions essential. Recent advancements in satellite remote sensing technology, including higher spatial resolution and sensitivity, have enhanced our ability to detect SO₂ emissions from volcanoes worldwide. However, traditional fixed-threshold algorithms struggle to automatically distinguish volcanic SO₂ emissions from non-volcanic sources. Additionally, accurately quantifying SO₂ emissions is challenging due to their dependence on plume height, particularly when reaching high altitudes. To address these challenges, we developed an Artificial Intelligence (AI) algorithm that detects and quantifies volcanic SO₂ emissions in near real-time. Our approach utilizes a Random Forest (RF) model, a supervised Machine Learning (ML) algorithm, to identify volcanic SO₂ emissions and integrates Cloud Top Height (CTH) data to enhance the accuracy of SO₂ mass quantification during intense volcanic eruptions. This AI algorithm, fully implemented in Google Earth Engine (GEE), leverages data from the TROPOspheric Monitoring Instrument (TROPOMI) aboard the Copernicus Sentinel-5 Precursor (S5P) satellite to automatically retrieve daily volcanic SO₂ plumes and CTH. We validated the model's performance against the Radius classifier, a state-of-the-art tool, and generalized its application across various volcanoes (Etna, Villarrica, Fuego, Pacaya, and Cumbre Vieja) with differing degassing activities, SO₂ emission rates, and plume geometries. Our findings demonstrate that the proposed AI approach effectively identifies and quantifies SO₂ plumes emitted by different volcanoes, enabling the investigation of SO₂ emission time series that reflect magma dynamics.