Assessing landslide susceptibility and dynamics at cultural heritage sites by integrating machine learning techniques and persistent scatterer interferometry

Landslides can significantly affect cultural heritage sites worldwide, often leading to irreversible damage and loss of invaluable cultural assets, and the assessment of the spatio-temporal distribution of such processes in culturally relevant sites is still a challenge. In this study, we propose a workflow to assess landslide susceptibility at the catchment scale and landslide dynamics, in terms of state of activity, at the slope scale with reference to built environments. A fully open-source and quantitative approach that integrates machine learning methods and persistent scatterer interferometry is proposed. The workflow was tested to identify cultural heritage sites potentially affected by landslides in a catchment of the Northern Apennines (Italy) characterized by the occurrence of earth slides and earth flows. The research reveals that 18 sites are located in highly susceptible terrains and five of them display notable displacement rates. Two sites in the highest susceptibility class and with high displacements rates were selected as case studies. One of the sites showed displacement rates up to 8 mm/year, while the second one up to 80 mm/year. A seasonal pattern of displacements was observed, with higher rates during summer and autumn. The analysis suggested a remarkable influence of topographic conditioning factors for the identification of earth slide susceptibility, while lithology was more important for the identification of earth flow susceptibility. Limitations due to the widespread occurrence of landslides characterized by a complex style of activity and the yearly update schedule of the interferometric data used are acknowledged. Nonetheless, the proposed workflow demonstrates its replicability with minimal operational costs to assess landslide susceptibility and state of activity in diverse geomorphological contexts.