|Paper title||Near-Real Time Artificial Intelligence Approach for Volcanic Eruptions Monitoring using SEVIRI Earth Observation Data|
|Form of presentation||Poster|
Most volcano observatories are nowadays heavily reliant on satellite data to provide time-critical hazard information. Volcanic hazards refers to any potentially dangerous volcanic process that can threaten people and infrastructure, such as lava flows and pyroclastic flows. During an explosive eruption, a major hazard to population can be represented by the ejection in the atmosphere of gases and ash, with the consequently creation of a volcanic plume, which can compromise aviation safety. Satellite remote sensing of volcanoes is very useful because it can provide data for large areas with a variety of modalities ranging from visible to infrared and radar. Satellite data suitable to monitor in near-real time the activity of a volcano are those acquired by the sensor Spinning Enhanced Visible and InfraRed Imager (SEVIRI), on board Meteosat Second Generation (MSG) geostationary satellite. SEVIRI has high temporal resolution (one image every 15 minutes) and good spectral resolution (12 spectral bands, including Visible, Near-Infrared and Infrared channels), providing a consistent amount of data exploitable for monitoring the eruptive activity of volcanoes. For example, Middle-Infrared (MIR) channels can be used to detect and quantify the thermal anomalies, whereas Thermal Infrared (TIR) bands can be adopted to observe and study volcanic clouds. Here, we propose a platform that exploits SEVIRI images to monitor in near real time the volcanic activity. In particular, we implemented an algorithm that detects the presence of volcanic thermal anomalies and, if they occur, measures the radiant heat flux to quantify these anomalies, checks if a volcanic plume appears and, consequently, uses machine learning algorithms to track the advancement of the plume and to retrieve its components (Figure 1).
SEVIRI data are downloaded automatically from the EUMETSAT DataStore using specific Python APIs; users can use the graphic interface of the platform to choose the time period of the images to download and to define the coordinates of the investigated region of interest. Once the SEVIRI images are downloaded, they are processed to detect the possible presence of volcanic thermal anomalies and, if so, the algorithm for the quantification of these anomalies and for the detection of a volcanic plume is started. Volcanic thermal anomalies are quantified by using a parameter called Fire Radiative Power (FRP) and, for each fire pixel detected, the FRP is calculated using the Wooster’s MIR radiance approach. The detection of a volcanic plume is performed exploiting the TIR bands of SEVIRI images: the brightness temperature difference (BTD) between bands at 10.8 µm and at 12.0 µm highlights the presence of thin volcanic ash, whereas the difference between bands at 10.8 µm and 8.7 µm emphasizes the presence of SO2. Starting from this consideration, a machine learning (ML) algorithm was developed to detect volcanic plumes and to retrieve their content of ash and SO2. This algorithm exploits manually labeled image regions to train a classifier that is able to recognize the plume and plume patches corresponding to ash, SO2 and mixing of ash and SO2. The learned classifier has the ability to generalize this approach and to classify automatically new images and all the newly emitted volcanic plumes. This near-real time approach for volcanic eruptions monitoring is daily applied to assess the status of Mt. Etna (Italy), but it can be applied successfully also to any other volcano covered by SEVIRI, just setting the correspondent coordinates in the graphic interface of the platform.