Work

Feature-based clustering of DInSAR time series to unveil complex landslide kinematic patterns

Year: 2026
Type: article
Abstract: Differential Interferometric Synthetic Aperture Radar (DInSAR) data processing generates displacement time series widely used for landslide mapping and monitoring. In many cases, displacement time series exhibit non-linear behaviour, which limits the applicability of conventional velocity-based classifications for detecting and interpreting landslide kinematics. Time series clustering has recently been introduced to identify displacement patterns. However, existing workflows often divide time series into a few and simple displacement styles, failing to discriminate between behaviours associated with gradual to abrupt velocity changes. This study investigates a clustering framework to unveil unknown displacement patterns in a large time series dataset. It exploits data dimensionality reduction by feature extraction to improve cluster mapping. A set of statistical features is proposed to describe different time series characteristics (e.g. average velocity, timing of changepoint, linear trend, and curvature), which is potentially useful for cluster-based classification of any DInSAR dataset. This study uses COSMO-SkyMed satellite data and tests the method over the historical town of Civitacampomarano (southern Italy), where a destructive landslide occurred in March 2017, causing the evacuation of a neighbourhood. Clustering helps identify displacement patterns and slope sectors requiring further investigation and distinguish between long-term ground instabilities and the ones potentially triggered by critical rainfall events or earthquakes. Particularly, the cluster of the March 2017 landslide was isolated, helping to constrain its spatial extent and investigate its triggering factors.
Language: English
Subfield: Geophysics
Open Access status: gold
Estimated APC paid: $1,790
OpenAlex ID: W7237340004
FWCI:
Cites: 43
Cited by:
Related to: