Detection and mapping of small-scale and slow-moving landslides from very high resolution optical satellite data

Abstract

Small slope failures are often ignored because of their perceived less severe impact. Although they may have small velocity, small slope failures can cause damages to facilities such roads and pipelines. The main objective of this research is to utilise very high resolution Pleiades-1 data to extract surface features and identify surface deformations susceptible to small slope failures. An algorithm was developed using object-based image analysis (OBIA), Pleiades-1 imagery, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) and Real Time Kinematic-Global Positioning System (RTK-GPS) data. Using the OBIA algorithm four different object attribute parameters namely spectral, textural, spatial and topographic characteristics were applied in a rule-based classification, for semi-automated detection of small translational landslides. The developed OBIA algorithm was further modified by using Pleiades-1 imagery, Nearest Neighbors (k-NN) and Support Vector Machine (SVM) techniques in example-based classification for the detection of small landslides, with focus on the effects of the training samples size and type on the results of the classification. The horizontal displacement of the landslides was investigated based on sub-pixel image correlation method using Pleiades-1 images and Shuttle Radar Topographic Mission (SRTM). Kalman filtering method and RTK-GPS observations from TUSAGA-Aktif Global Navigation Satellite System (GNSS) Network in Turkey were utilised to formulate kinematic analysis model for the landslides. The developed algorithms were validated in Kutlugün test site in Northeastern Turkey. In the rule-based classification results, a total of 123 small landslides covering a total area of approximately 413.332 m2 were detected. The size of landslides detected varied between 0.747 and 7.469 m2. The detected landslides yielded user’s accuracy of 81.8%, producer’s accuracy of 80.6%, quality percentage of 82% and computed kappa index of 0.87. In the small landslides detection using the example-based classification, the SVM method had higher producer accuracy (85.9%), user accuracy (89.4%) and kappa index (0.82) compared to the k-NN algorithm that had producer accuracy (83.1%), user accuracy (86.0%) and kappa index (0.80). A total of 128 small landslides were detected using k-NN algorithm, while a total of 134 landslides were detected using SVM algorithm. The displacement results from RTK-GPS measurements varied from 2.77 mm to 24.87 mm in 6 months, while the velocities varied from 0.80 mm to 8.28 mm/6 month. The displacements from optical image correlation agreed well with RTK-GPS results and provided a more uniform movement pattern than could be derived solely using the RTK-GPS measurements. The landslide movements are dominantly toward the north direction. These trends agree with the results of previous study in the area. The main contributions of this research include – development of a comprehensive metrics to quantify the attribute parameters of small landslides, derivation of susceptibility and inventory maps for small landslides, and the design of an early warning system for small slope failures on highway infrastructures. The results of this research will add to the increasing applications of Pleiades-1 image in landslide investigations.