An improved multi-model risk assessment of creep-type landslide dynamics with multi-temporal synthetic aperture radar (MT-InSAR)

Creep-type landslides in high-mountain regions threaten ecosystems, communities and infrastructure due to a combination of several factors, including topography, glacial deposits and sensitive geological formations. Accurately monitoring these landslides is challenging due to complex, nonlinear causal relationships and limitations in capturing multiscale spatio-temporal dynamics. To address these challenges, this study develops an improved multi-model framework for landslide risk assessment that integrates Multi-Temporal Interferometric Synthetic Aperture Radar (MT-InSAR) from multi-orbit Sentinel-1 data (October 2014–May 2024) with joint persistent scatterer (PS) and distributed scatterer (DS) processing, multiscale geographically weighted regression (MGWR), K-medoids clustering and wavelet analysis. Also, hydro-climatic time-series variables, including soil moisture content (SMC), precipitation, snow depth and soil temperature, and terrain-derived geomorphometric factors (e.g., slope, Topographic Wetness Index [TWI] and irrigation channel distance) are included. This framework was implemented in the Parsan Valley in the Hindukush–Himalayan–Karakoram Range (HHK) region of Pakistan. The average line-of-sight (LOS) deformation results revealed a significant movement (−60 to 60 mm/year), concentrated in the central and south-eastern valley. Wavelet analysis identified significant SMC and deformation correlations (Spearman's ρ = −0.66, p < 0.001) with lags of 2–142 days, reflecting seasonal influences and suggesting a complex interplay with several environmental processes. MGWR (adjusted R² = 0.676) highlighted TWI (β = 0.580) and irrigation channel distance (β = 0.584) as primary drivers, with slope (β = 0.063) as a secondary factor, driven by a dual moisture regime of monsoon rainfall (June–September), spring snowmelt (April–June) and spring-fed irrigation channels, triggering shallow, translational sliding. K-medoids clustering delineated high-risk zones, with an R² value of 0.75 between deformation and slope angle. This framework improves landslide monitoring and risk assessment, supporting targeted mitigation with potential application in other areas.