Interpretación visual y digital de datos de sensores remotos para la identificación de deslizamientos rotacionales y traslacionales

Abstract

Esta tesis presenta un método para la detección y ubicación de movimientos en masa al analizar su distribución, patrón y recurrencia. El método propuesto para la detección de movimientos en masa utiliza herramientas de la geomática buscando reducir tiempos y facilitar la creación de inventarios de movimientos en masa. En este estudio se realizó la identificación visual de rasgos pictórico - morfológicos en deslizamientos ya identificados, y su posterior uso para la definición de criterios de clasificación de deslizamientos en un método semiautomático de clasificación basado en objetos geográficos (GEOBIA). Se identificaron patrones cualitativos que identifican los rasgos pictórico – morfológicos de deslizamientos. Posteriormente se realizó la clasificación basada en objetos, generando la segmentación de la imagen seguido de la clasificación basada en objetos geográficos identificando las coberturas de la tierra y deslizamientos. A partir de los patrones cualitativos se refinaron los objetos clasificados como deslizamientos y se clasificaron mediante el uso de la curvatura del terreno en deslizamientos del subtipo traslacional o rotacional. El resultado se validó en términos de área entre los polígonos clasificados como deslizamientos y de un inventario de movimientos en masa precedente. Se obtuvo que el área correctamente clasificada se situó entre un 60% a 50% y el área erróneamente clasificada fue entre el 25% a 15%. (Texto tomado de la fuente)

The detection and location of mass movements allows to analyze their distribution, pattern and recurrence. The creation of methods that use tools provided by Geomatics seeks to reduce times, facilitate the creation, production of inventories and mass movement maps; which are the basic input for the generation of susceptibility maps. Computer advances allow the use of tools for the semi-automatic location of objects in satellite images, although there are several studies on the use of geographic information systems for the detection of these natural events, in Colombia a methodology has not been implemented or studied efficient and economical, which is an opportunity to further develop the implementation of geomatics in the location of mass movements in large areas. This project proposes the visual identification of pictorial-morphological features in already identified landslides, and their subsequent use for the definition of landslide classification criteria in a semi-automatic classification method based on geographic objects (GEOBIA). This semi-automatic method identifies some types of landslides with the use of satellite imagery supported by an existing inventory of mass movements. The method was applied in (2) two areas located in the rural part of the municipality of Villavicencio in the department of Meta, using satellite multispectral images from the Sentinel 2 mission and a digital terrain model (DTM) obtained from radar images of the Sentinel mission 1. In a first step, qualitative patterns were identified that identify a pictorial-morphological feature in landslides of an existing inventory of mass movements. For this, spectral criteria, temporal criteria, spatial criteria and an area criterion were used. Subsequently, the classification based on objects was carried out, generating the segmentation of the image followed by the classification based on geographical objects, identifying the land covers and landslides located in the study area. Next, with the identified qualitative patterns, the use of parameters such as the normalized vegetation index (NDVI), the soil gloss index (S2 BI), contextual data and the slope of the terrain was defined, which allowed to refine the objects. classified as landslides obtained from GEOBIA. Once these polygons were refined with the curvature of the terrain, they were classified into landslides of the translational and rotational subtype. Finally, the result obtained was validated in terms of area (extension) between the polygons classified as landslides and the pre-existing mass movement inventory data. It was obtained that the correctly classified area was between 56.6 % and 51 % in the two study areas analyzed. Regarding the erroneously classified area, 17 % and 25 % were obtained. According to the results obtained from this methodology, these allow an approximation of delimitation of candidate areas for the presence of landslides in large areas.