EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Semi-automated regional classification of slow rock slope deformations integrating kinematics, activity and spatial complexity

Chiara Crippa, Federico Agliardi, Paolo Frattini, Margherita C. Spreafico, Giovanni B. Crosta, and Elena Valbuzzi
Chiara Crippa et al.
  • Università Milano Bicocca, DISAT_CSS1, Departement of Earth and Environmental Science, Milano, Italy (

Slow rock slope deformations are widespread in alpine environments. They affect giant volumes and evolve over thousands of years by progressive failure, resulting in long-term slow movements threatening infrastructures and potential evolution into massive collapses. In the alpine sector of Lombardia (Italian Central Alps), 208 mapped slow rock slope deformations affect a total area exceeding 580 km2 and interact with a variety of elements at risk including settlements, hydroelectric facilities and lifelines characterized by different vulnerability to both slow and progressive deformations. In this context, a systematic, reliable and cost-effective approach is required to classify slow rock slope deformations on the regional scale for landplanning, prioritization and analysis of interactions with elements at risk, depending on their style of activity, including not only mean deformation rate, but also their kinematics and spatial complexity. In this work, we implemented a toolbox that integrates different approaches to classify a large dataset of slow rock slope deformations in discrete groups, according to the deformation style and morpho-structural expression of individuals, mapped on regional scale and characterized through remote sensing techniques. The landslide dataset used in this study was obtained by a “semi-detail” geomorphological and morpho-structural mapping on aerial imagery and DEM, performed on regional scale yet including local-scale information (e.g. tectonic lineaments, morpho-structures, landforms, nested deep-seated landslides) and a full set of geological and morphometric attributes. To characterize landslide activity, we use Persistent-Scatterer Interferometry (PSI) data, including PS-InSARTM and SqueeSARTM acquired by different sensors (ERS, Radarsat, Sentinel 1A/B) over different time periods from 1992 to 2017. Since Line-of-Sight velocity of point like data can hamper a correct evaluation of both landslide kinematics and deformation rates, for each phenomenon we automatically selected the most complete PSI datasets. From these, through a 2DSAR decomposition procedure, we derived 2D velocity components and computed the magnitude and orientation of the 2D total displacement vector T.  We then applied a supervised machine learning procedure to automatically classify the kinematics of each landslide (i.e. translational, roto-translational, rotational) depending on the statistical distribution of the T vector orientation. As the evaluation of a representative landslide mean deformation rate is strongly affected by spatial heterogeneity and landslide mass segmentation, we implemented an original peak analysis of the velocity distribution in each landslide to calculate a modal velocity of the main body and automatically outline nested sectors with differential displacement rates. Finally, we classified landslides in types, representative of different styles of activity and potential interaction with elements at risk, by combining PSI analysis results with geological, morpho-structural and morphometric variables in a multivariate statistical analysis framework including sequential Principal Component and K-medoids Cluster Analysis. The entire analysis workflow runs in a semi-automated way through a set of GIS and MatlabTM tools. Our procedure can be applied to different large landslide datasets, providing a fast and cost-effective support to landslide classification, risk analysis, landplanning and prioritization of local-scale studies aimed at granting safety and infrastructure integrity.