What question did this study set out to answer?

To characterize and predict slow-moving landslide displacements using InSAR and ensemble regression models.

February 19, 2026Natural Hazards0 citationsOpen Access

Displacement prediction of slow-moving landslides using InSAR and ensemble regression models based on slope units

SMSandra Lucía Cobos MoraUniversidad Católica de Cuenca VRVíctor Rodríguez‐GalianoUniversidad de Sevilla ZFZhice FangChina University of Geosciences

Key Points

To characterize and predict slow-moving landslide displacements using InSAR and ensemble regression models.
Utilized InSAR with SBAS technique for ground deformation analysis in an Andean region.
Aggregated LOS displacement time series from 2021 to 2023 using mean and extreme value methods.
Decomposed time series into trend and periodic terms using static and dynamic predictors modeled by Extreme Gradient Boosting.
Identified three zones of slow-moving deformation with LOS velocities ranging from −68 to 388.6 mm/year.
Achieved RMSE values of 0.81 mm for trend term in Zone 1, indicating high predictive accuracy.
Highlighting the dominant role of groundwater storage and precipitation in influencing displacement trends.

Abstract

Ground displacement is a key indicator of slope instability and plays a crucial role in mitigating landslides triggered by climate-related factors. Interferometric Synthetic Aperture Radar (InSAR) has become an essential tool for detecting and characterizing large-scale, slow-moving deformations. This study (i) characterizes ground deformation in an Andean region with known landslide activity using the Small Baseline Subset (SBAS) InSAR technique, and (ii) proposes a novel predictive framework for slow-moving displacements. Line-of-Sight (LOS) displacement time series (TS) from 2021 to 2023 were aggregated at the slope-unit scale based on mean and extreme values. Each TS was decomposed into trend and periodic terms and described using static and dynamic predictors, the latter computed over 7–28-day intervals. Both components were modeled using Extreme Gradient Boosting (XGBoost). InSAR-based characterization of the study area identified three zones exhibiting slow-moving deformation, with LOS velocities ranging from − 68 to 388.6 mm/year (ascending) and − 245.7 to 165.1 mm/year (descending). The predictive framework achieved high accuracy, particularly when using mean-based TS (MeanDts) for both terms. For the trend term, MeanDts achieved RMSE values of 4.18–4.74 mm, MAE values of 1.94–2.53 mm, and R 2 ≥ 0.98. For the periodic term, RMSE ranged between 2.30 and 2.36 mm, MAE averaged 1.61 mm, and R 2 reached 0.98. In contrast, the maximun-based series (MaxSignDts) series showed a marked performance decline for the periodic term. Model performance was highest in Zone 1, which includes the 2023 Causal - Alausí landslide. In this area, MeanDts asc achieved RMSE = 0.81 mm, MAE = 0.61 mm, R 2 = 0.999, Max Err = 2.35 mm, and Mean Err = 0.61 mm for the trend term; and RMSE = 0.88 mm, MAE = 0.64 mm, R 2 = 0.998, Max Err = 2.62 mm, and Mean Err = 0.64 mm for the periodic term. Analysis of predictor importance revealed that precipitation, elevation, aspect, and proximity to faults influenced the trend term, with groundwater storage (Gws) exerting the strongest effect. For the periodic term, Gws remained the dominant driver, while precipitation contributed marginally. Overall, mean-based TS consistently outperformed maximum-based ones, with no consistent advantage observed between ascending and descending geometries. This framework provides spatially explicit, long-term deformation forecasts that can inform risk-based planning and the design of climate-resilient infrastructure in landslide-prone mountain regions.

Ask AI

Helpful

Bookmark

View Full Paper