Volcanic cloud detection using Sentinel-3 satellite data by means of neural networks: the Raikoke 2019 eruption test case

• Main Authors: Petracca, Ilaria, Santis, Davide, Picchiani, Matteo, Corradini, Stefano, Guerrieri, Lorenzo, Prata, Fred, Merucci, Luca, Stelitano, Dario, Frate, Fabio, Salvucci, Giorgia, Schiavon, Giovanni
• Format: Text
• Language: English
• Published: 2022
• Subjects: Eyjafjallajökull
• Online Access: https://doi.org/10.5194/amt-2022-173; https://amt.copernicus.org/preprints/amt-2022-173/

The accurate automatic volcanic cloud detection by means of satellite data is a challenging task and of great concern for both scientific community and stakeholder due to the well-known issues generated by a strong eruption event in relation to aviation safety and health impact. In this context, machine learning techniques applied to recent spaceborne sensors acquired data have shown promising results in the last years. This work focuses on the application of a neural network based model to Sentinel-3 SLSTR (Sea and Land Surface Temperature Radiometer) daytime products in order to detect volcanic ash plumes generated by the 2019 Raikoke eruption. The classification of the clouds and of the other surfaces composing the scene is also carried out. The neural network has been trained with MODIS (MODerate resolution Imaging Spectroradiometer) daytime imagery collected during the 2010 Eyjafjallajökull eruption. The similar acquisition channels of SLSTR and MODIS sensors and the events comparable latitudes foster the robustness of the approach, which allows overcoming the lack in SLSTR products collected in previous mid-high latitude eruptions. The results show that the neural network model is able to detect volcanic ash with good accuracy if compared with RGB visual inspection and BTD (Brightness Temperature Difference) procedure. Moreover, the comparison between the ash cloud obtained by neural network and a plume mask manually generated for the specific SLSTR considered images, shows significant agreement. Thus, the proposed approach allows an automatic image classification during eruption events, which it is also considerably faster than time-consuming manually algorithms (e.g. find the best BTD product-specific threshold). Furthermore, the whole image classification indicates an overall reliability of the algorithm, in particular for meteo-clouds recognition and discrimination from volcanic clouds. Finally, the results show that the NN developed for the SLSTR nadir view is able to properly classify also the SLSTR ...