Detection and Prediction of Landslide Vulnerability through Case Study using DInSAR Technique and U-net Model

Landslides are among the deadliest natural catastrophes because they usually occur after other severe disasters, causing enormous damage to already vulnerable systems. In recent years remote sensing methodologies have grown prominence in landslide monitoring and prediction due to their ability to mitigate hazards. In remote sensing, the Interferometry technique is employed in Synthetic Aperture Radar. In this methodology, the DInSAR technique is used for the phase difference calculation over a specific area using the sensors of the Sentinel satellites. SR-530 landslide is the chosen field of study for assessing the properties and parameters required to detect and predict landslides. The Digital Elevation model can be generated from the Sentinel satellite imagery using the DInSAR technique. Apart from DEM, Slope and Normalized Difference Vegetation Index are also used to detect and quantify the presence of land deformation and erosion. Natural color (RGB) image is also considered in landslide detection. These parameters are used to predict the activity of landslides using the U-net Model through training on various data samples across the globe. Immediate recognition can be made once the landslide starts showing its signature like land deformations, changes in erosion, vegetation, etc. From the obtained results, forthcoming landslides can be detected.