Method for landslides detection with semi-automatic procedures: The case in the zone center-east of Cauca department, Colombia
Landslides are a common natural hazard that causes human casualties, but also infrastructure damage and land-use degradation. Therefore, a quantitative assessment of their presence is required by means of detecting and recognizing the potentially unstable areas. This research aims to develop a metho...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2020 |
| País: | Colombia |
| Institución: | Universidad Nacional de Colombia |
| Repositorio: | Repositorio UN |
| Idioma: | inglés |
| OAI Identifier: | oai:repositorio.unal.edu.co:unal/77975 |
| Acceso en línea: | https://repositorio.unal.edu.co/handle/unal/77975 |
| Access Level: | acceso abierto |
| Palabra clave: | 550 - Ciencias de la tierra 600 - Tecnología (Ciencias aplicadas) 620 - Ingeniería y operaciones afines movimientos en masa inventario de deslizamientos detección semi-automática de deslizamientos teledetección random forest landslide mass movement landslide inventory, semi-automatic detection of landslides remote sensing |
| Sumario: | Landslides are a common natural hazard that causes human casualties, but also infrastructure damage and land-use degradation. Therefore, a quantitative assessment of their presence is required by means of detecting and recognizing the potentially unstable areas. This research aims to develop a method supported on semiautomatic methods to detect potential mass movements at a regional scale. Five techniques were studied: Morphometry, SAR interferometry (InSAR), Persistent Scatterer InSAR (PS-InSAR), SAR polarimetry (PolSAR) and NDVI composites of Landsat 5, Landsat 7, and Landsat 8. The case study was chosen within the mid-eastern area of the Cauca state, which is characterised by its mountainous terrain and the presence of slope instabilities, officially registered in the CGS-SIMMA landslide inventory. This inventory revealed that the type `slide' occurred with 77.4% from the entire registries, `fall' with 16.5%, followed by `creeps' with 3%, flows with 2.6%, and `lateral spread' with 0.43%. As a result, we obtained the morphometric variables: slope, CONVI, TWI, landform, which were highly associated with landslides. The effect of a DEM in the processing flow of the InSAR method was similar for the InSAR coherence variable using the DEMs ASTER, PALSAR RTC, Topo-map, and SRTM. Then, a multiInSAR analysis gave displacement velocities in the LOS direction between -10 and 10 mm/year. With the dual-PolSAR analysis (Sentinel-1), VH and VV C-band polarised radar energy emitted median values of backscatters, for landslides, about of -14.5 dB for VH polarisation and -8.5 dB for VV polarisation. Also, L-band fully polarimetric NASA-UAVSAR data allowed to nd the mechanism of dispersion of CGS landslide inventory: 39% for surface scattering, 46.4% for volume dispersion, and 14.6% for double-bounce scattering. The optical remote sensing provided NDVI composites derived from Landsat series between 2012 and 2016, showing that NDVI values between 0.40 and 0.70 had a high correlation to landslides. In summary, we found the highest categories related to landslides by Weight of Evidence method (WofE) for each spaceborne technique applied. Finally, these results were merged to generate the landslide detection model by using the supervised machine learning method of Random Forest. By taking training and test samples, the precision of the detection model was of about 70% for the rotational and translational types. |
|---|