Volcanic clouds detection applying machine learning techniques to GNSS radio occultations

Abstract

AbstractVolcanic clouds detection is a challenge especially when meteorological clouds are present in the same area. Several algorithms have been developed to detect and monitor volcanic clouds by using satellite instruments based on different remote sensing techniques. This work aims at classifying volcanic clouds based on atmospheric profiles retrieved by the GNSS (Global Navigation Satellite Systems) radio occultation technique. We collocated the radio occultations with the volcanic cloud detection from AIRS (Atmospheric InfraRed Sounder) and IASI (Infrared Atmospheric Sounding Interferometer) for 11 big eruptions happening in the period 2008–2015 resulting in about 15000 profiles. We created an archive with the collocations and a corresponding number of profiles in “non-volcanic” environment in the same area and on the same period of the year. A support vector machine algorithm was applied to the archive in order to classify the clouds and to distinguish the volcanic clouds from the other types. The model performances are promising: the GNSS radio occultations are able to distinguish the volcanic clouds with an accuracy higher than 80% when the eruption occurs at high latitudes. The performances of the model are affected by the number of collocations used for the training. Nowadays, the number of radio occultations is higher than in the period considered in this research, making this work a pioneering study for a future operational product.