Exploring clustering methods for the analysis of InSAR time series data: a case study of landslides in the Palos Verdes Peninsula, California, USA

Abstract

Landslides are a persistent hazard along Palos Verdes Peninsula, California, USA, where complex terrain and urban development necessitate effective monitoring of ground deformation. This study investigates the application of unsupervised clustering technique to Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) time series data to classify displacement behaviours and support regional landslide assessment. Principal Component Analysis (PCA) and K-means clustering were applied to ascending and descending Line-of-Sight (LOS) measurements, followed by decomposition into vertical and slope-aligned horizontal components. The clustering outputs were compared against known landslide complexes, such as Portuguese Bend, Abalone Cove, and Klondike Canyon, revealing strong spatial agreement and enabling broader interpretation across the peninsula. Three kinematic behaviour types were defined based on combinations of horizontal and vertical cluster patterns, allowing differentiation between kinematically stable regions, active landsliding, and ambiguous zones. The results demonstrate that the clustering framework is time-efficient, scalable, and capable of highlighting both known and previously unreported deformation patterns. However, the approach has limitations due to coherence loss in highly active zones, assumptions made in LOS decomposition, and the potential influence of non-landslide processes such as seasonal change or anthropogenic activity. Despite these constraints, the method provides a practical and physically interpretable model for regional landslide screening, offering insights that support hazard identification and long-term monitoring across complex, landslide-prone terrain.