Use of interferometric satellite radar data for the improvement of space-time prediction models of landslides

The increasing impact of natural and human-induced disastrous phenomena, further intensified by climate change, highlights the need to understand, analyse and predict these phenomena. Landslides, in particular, represent one of the most catastrophic events that annually cause significant economic damage and human loss worldwide. In recent decades, there has been a great revolution in the study of these phenomena, thanks to the introduction of remote sensing techniques, and particularly the advent of Synthetic Aperture Radar (SAR) sensors in the 1990s. These active sensors operate independently of lighting and weather conditions, acquiring images that, after being processed through interferometric techniques (InSAR), provide a consolidated tool for studying ground deformations, including landslides. Such technologies offer valuable support to administrative entities and civil protection authorities for risk management and reduction. In this context the present thesis work is situated, mainly carried out at the Department of Earth Sciences of the University of Florence, and enriched by collaboration with foreign institutions and participation in the European research project RASTOOL (European ground motion risk assessment tool). The latter, particularly, has provided a set of tools for the implementation of automatic techniques for analysing InSAR data, which constitute one of the main focuses of the thesis work. Automatic and semi-automatic techniques are becoming increasingly necessary within the scientific community to efficiently handle and exploit the vast amount of radar data that is currently available. This need has become even more evident since the launch of the Sentinel-1 (S1) constellation and the development of the European Ground Motion Service (EGMS). Specifically, EGMS is the first service worldwide to provide continental-scale ground motion data derived from S1 with open access and millimetre accuracy. The large amount of data and the scale of application, which ranges from regional to national in the performed activities, have required the employment of automatic methods such as clustering and machine learning (ML) techniques. The first approach developed in this thesis concerns the assessment of landslide susceptibility through a ML model that integrates satellite radar data, with applications at both regional and national scales. The second approach employs clustering methods, ML techniques, and geomorphometric analyses for identifying and classifying areas of active deformation on regional and national scales by using S1 data. Finally, the third approach focuses on the use of ML algorithms to identify areas where accelerations or decelerations of ground movement, identified by S1 data, are more likely to occur. The approaches proposed in this thesis thus highlight the advantages and limitations of using satellite radar data in landslide analysis, demonstrating how InSAR semi-automatic techniques can constitute valuable tools for space-time managing and mitigating risks associated with such phenomena.