Classifying Mt. Etna Lava Flows using PlanetScope Image and U-Net-based Deep Learning

Mt. Etna was in an eruption period between 2017 and 2019, which was represented by several lava flows emitted from summit craters and fissures on its flank. Considering the hazards of lava flows, we classified the lava flow emplacement area by exploiting high-resolution PlanetScope images combined with deep learning model using U-Net architecture. A posteruption images that properly captured lava flows due to December 2018 eruption were selected for classification. A postclassification process was applied to filter out the falsely classified pixels. Confusion matrix analysis was used to evaluate the model performance and improve the accuracy of the classification results. The results show that the deep learning models produced overall accuracies of about 76% with Kappa coefficient of about 0.58. This finding provides an acceptable and simple method for mapping lava flows and would be useful in hazard assessment given that there are several tourist places and densely populated areas on the flanks of Mt. Etna.