Dynamics of slow-moving landslides in the Eastern Cordillera of the Central Andes derived from optical satellite imagery

The Eastern Cordillera of the Central Andes in Northwestern Argentina and Southern Bolivia is an actively deforming mountain front with elevations ranging from ~1 km in the foreland to ~4 km and higher on the Central Andean Plateau. The orographic barrier induces a strong climatic and environmental east-west gradient with peak rainfall in the steep eastward-facing slopes. Frequent rainstorms during the South American summer monsoon in combination with fault-weakened lithologies drive mass-movement processes. The resulting debris flows and landslides pose a serious threat to the local infrastructure.

In this study, we integrate satellite-based optical remote sensing data over the last 10 years to characterize the long-term dynamics of slow-moving landslides in the eastern Central Andes in Argentina. In this way, we aim to establish potential relationships between climatic seasonality, seismic activity and landslide deformation signals. We apply a combination of pixel and feature-based tracking approaches to a data set comprising a network of medium-resolution Sentinel-2 and Landsat 8, and high-resolution SPOT7 optical images. The final ground displacement time series in east-west and north-south directions were reconstructed through time-series inversion. Vertical variations were obtained by comparing high-resolution Digital Surface Models (DSM) produced from tri-stereo SPOT7 images. We attempt to improve the detection of very slow-moving landslides with velocities below 0.5 m/yr by stacking multiple matching pairs and relying on feature-based tracking approaches. In some examples, the displacement time series reveal metric ground displacements following earthquake events observed in the region, changing the dynamics of the landslide.

This study emphasizes the usefulness of large-scale, decadal-long time series of optical satellite imagery and presents a novel GPU-based approach of combining computer vision feature tracking methods with classic correlation based block matching.