Use of Sentinel-2 images for the detection of precursory motions before landslide failures

Abstract

The Sentinel-2 optical satellites provide a global coverage of land surfaces with a 5-day revisit time at the Equator. We investigate the ability of these freely available optical images to detect precursory motions before rapid landslides. A 9-month time-series of displacement is derived from Sentinel-2 data over a major landslide in the French Alps, which exhibited a sudden reactivation in June 2016. This analysis reveals a 7-month period of low activity (≤1 m), followed by a sudden acceleration of 3.2 ± 1.2 m in 3 days, before the failure of a mass of about 2 to 3.6 106 m3. The location of this precursory motion is consistent with that of the slow motions occurring since 2001 (about 1 m/year), as revealed by aerial photographs and LiDAR analysis. This change in activity over a very short period of time (days) emphasizes the value of the frequent revisit time of Sentinel-2, despite its medium resolution of 10 m. We finally simulate the ability of Sentinel-2 for detecting these precursory patterns before a rapid landslide occurs, based on typical Voight's laws for creeping materials, characterized by a power law exponent α. Based on this analysis and on global cloud cover maps, we compute the probability to detect pre-failure motions of landslides using the Sentinel-2 constellation. This probability is highly heterogeneous at the global scale, affected by the revisit time of the satellite and the cloud cover. However the main factors controlling this detection ability are the properties of the landslide itself (its size and the α parameter), with almost 100% of detection probability for α = 1.3 and 0% for α = 1.8. Despite all these limitations, probability to detect a motion before a landslide failure often reaches 50% for classical landslide parameters. These results open new perspectives for the early warning of large landslide motion from global and open source remote sensing data.