Landslide Suspectibility Mapping Using Hybrid Deep Learning

To put it simply, landslides are the collapse of a slope’s worth of land, posing a threat to human, animal, and man-made life under varying and often erratic climatic and lithological conditions. The development of cutting-edge space technology has allowed for the expansion of synthetic aperture radar (SAR) interferometry in the face of disaster. Copernicus Sentinel 1 SAR data products, with a temporal resolution of 12 days, are freely available, enriching periodic monitoring of the Earth’s surface. Over the course of several decades, differential SAR interferometry (DInSAR) techniques have been widely used for the purpose of tracking and identifying surface distortion. Over 105 landslides occurred in the Kodagu district of Karnataka during the 15th and 17th of August 2018. Before and after landslide occurrences, Sentinel-1 datasets acquired in Interferometric Wide Swath (IW) mode are utilised. Topographic and atmospheric inaccuracies have a significant impact on the displacement result derived from DInSAR. Due to its non-uniform accuracy variance, DEMs must be evaluated prior to being used for a variety of applications. DEMs and InSAR produced DEMs are evaluated with respect to their vertical and horizontal accuracy using Survey of India (SOI) toposheets as a standard of comparison. After considering their accuracy in both the vertical and horizontal planes, researchers have concluded that ALOS are the best option for topographic phase removal. Use of ALOS for InSAR analysis over the Kodagu district is recommended as it shows the least amount of error compared to other DEMs. Sentinel 1 can be utilised for assessment of larger landslides, and it is recommended to use corner reflectors to produce promising findings, according to a time series analysis done across the selected landslide regions using the Hybrid Deep Learning approach.