This paper presents a novel method of landslide detection by exploiting the Mask R-CNN capability of identifying an object layout by using a pixel-based segmentation, along with transfer learning used to train the proposed model. A data set of 160 elements is created containing landslide and non-landslide images. The proposed method consists of three steps: (i) augmenting training image samples to increase the volume of the training data, (ii) fine-tuning with limited image samples, and (iii) performance evaluation of the algorithm in terms of precision, recall, and F1 measure, on the considered landslide images, by adopting ResNet-50 and 101 as backbone models. The experimental results are quite encouraging as the proposed method achieves Precision equals to 1.00, Recall 0.93, and F1 measure 0.97, when ResNet-101 is used as backbone model, and with a low number of landslide photographs used as training samples. The proposed algorithm can be potentially useful for land-use planners and policymakers of hilly areas where intermittent slope deformations necessitate landslide detection as a prerequisite before planning.
1 January 2021 in Geomatics, Natural Hazards and Risk
Landslide susceptibility maps (LSMs) are generally prepared by integrating multiple prominent thematic layers, including DEM derived products (elevation, slope, and aspect), and other parameters such as lithology, geomorphology, LULC, etc. These parameters can be assigned optimum weights using the analytic hierarchy process (AHP) method, followed by a GIS-based weighted overlay analysis. In recent years, multi-temporal interferometric synthetic aperture radar (MT-InSAR) techniques have been rigorously explored, for land deformation detection and monitoring, by extracting highly stable measurement pixels using tens of SAR acquisitions simultaneously. In this research work, a GIS-based multi-criteria decision analysis to prepare LSMs is proposed, with MT-InSAR derived displacement estimates used as a critical input parameter. An LSM is generated by processing 20 ERS-1/2 and Envisat ASAR images, acquired over ∼120 sq. km wide river basin, located in Uttarakhand, India. The generated LSM is found to be congruent with the susceptible maps made available by the Geological Survey of India (GSI) under the National Landslide Susceptibility Mapping (NLSM) program. Preliminary results indicate that the majority of the unstable zones along the Alaknanda River are correctly identified. The approach is further implemented to generate an updated susceptibility map using 60 scenes of freely available Sentinel-1A dataset, followed by validation through actual field survey. This resulted in the generation of an updated susceptibility map, which helped in the identification of 44.5% new landslide susceptible zones (LSZs). Furthermore, the status of previously identified zones is also quantified. The performance of the proposed approach suggests its usability in generating and updating near-real-time LSMs.
A robust geodetic framework comprising Terrestrial Laser Scanner (TLS), Global Navigation Satellite Systems (GNSS), Robotic Total Station (RTS) and Multi-temporal InSAR (MT-InSAR) was employed first in India to investigate a landslide-prone Sirobagarh region, Uttarakhand at different spatial extents, and to evaluate the relationship amongst the displacement estimates obtained from the applied surveying techniques. TLS derived digital elevation models indicated displacements >5 m on the landslide upper scarp. GNSS and RTS based observations showed horizontal movements towards the Alaknanda river in the landslide slope direction (maximum values: 0.1305 and 0.045 m respectively), and downward vertical motion (largest subsidence magnitude: -2.1315 and -0.030 m respectively). MT-InSAR processing of Sentinel-1a images identified 21071 measurement pixels, highlighting subsidence around the landslide (mean velocity range: -0.110 to 0.008 m/yr). Analysis of displacement vectors using vector equality, cross-covariance, cross-correlation and Principal Component Analysis reveals that GNSS vertical displacement estimates were partially correlated with MT-InSAR measurements (correlated for epoch difference 2-3), whereas there was good cross-correlation between MT-InSAR and LiDAR observations throughout. The displacement estimates and their analyses evident unstable movement of the landslide scarp occurring due to debris flow and rainfall, and a relatively moderate subsidence activity in the surrounding areas lying in the landslide zone.
5 February 2016 in International Journal of Remote Sensing
This research compares two time-series interferometric synthetic aperture radar (InSAR) methods, namely persistent scatterer SAR interferometry (PS-InSAR) and small baseline subset (SBAS) to retrieve the deformation signal from pixels with different scattering characteristics. These approaches are used to estimate the surface deformation in the L’Aquila region in Central Italy where an earthquake of magnitude Mw 6.3 occurred on 6 April 2009. Fourteen Environmental Satellite (ENVISAT) C-band Advanced Synthetic Aperture Radar (ASAR) images, covering the pre-seismic, co-seismic, and post-seismic period, are used for the study. Both the approaches effectively extract measurement pixels and show a similar deformation pattern in which the north-west and south-east regions with respect to the earthquake epicentre show movement in opposite directions. The analysis has revealed that the PS-InSAR method extracted more number of measurement points (21,103 pixels) as compared to the SBAS method (4886 pixels). A comparison of velocity estimates shows that out of 833 common pixels in both the methods, about 62% (517 pixels) have the mean velocity difference below 3 mm year−1 and nearly 66% pixels have difference below 5 mm year−1. It is concluded that StaMPS-based PS-InSAR method performs better in terms of extracting more number of measurement pixels and in the estimation of mean line of sight (LOS) velocity as compared to SBAS method.