Assessing the elements at risk in volcanic areas by combining deep convolutional neural network and multispectral satellite images

Volcanic eruptions are spectacular but dangerous phenomena. Depending on their magnitude and location, they also have the potential for becoming major social and economic disasters. Some of the most important volcanic events include ash fallout, lava flows, and related phenomena, such as volcanic debris avalanches and tsunamis. The ongoing demographic congestion around volcanic structures, such as Mount Etna, increases the potential risks and costs that volcanic eruptions represent and leads to a growing demand for implementing effective risk mitigation measures. To fully evaluate the potential damage and losses that a volcanic eruption disaster may cause, the distribution and characterization of all the exposed elements must be considered. Over the past decades, advances in satellite remote sensing and geographic information system techniques have greatly assisted the collection of land cover data. However, assessment of the elements at risk is a lengthy and time-consuming process. In fact, usually data including all exposed elements and land uses are gathered from several Institutional web portals and very high-resolution satellite imagery, not freely available, manipulated by operators. Here, we propose a deep learning approach to automatically identify the elements at risk in high spatial resolution satellite images. In particular, a Convolutional Neural Network (CNN) model is adopted to classify land use and land cover in volcanic areas thus allowing to carefully assess the total exposure by using freely available satellite images. A retrospective analysis is conducted on Mount Etna highlighting changes in the exposure over the last decade.