Digital Terrain Models Derived from Unmanned Aerial Vehicles and Landslide Susceptibility

Abstract

Digital Terrain Models (DTMs) are among the most important spatial information tools used in geomorphological landslide assessment because they allow the extraction of crucial attributes, such as landslide geometry, slope, terrain curvature, etc. However, at a local scale, the assessment of remote volcanic terrains is difficult because the DTMs have poor spatial and temporal representation. Worldwide, geomorphological analysis of landslides processes in mountainous terrains with difficult access has benefited with virtual topography representations using high-resolution Digital Surface Models (DSMs) generated by imagery captured by unmanned aerial vehicles (UAV). These DSMs include not only the ground topography, but also other landscape elements such as vegetation, buildings, cars, etc. These natural and anthropogenic elements are considered as non-relevant information or noise to obtain only the ground information. Photogrammetric post-processing of the DSM is required to derive a DTM that represent only ground topography. This research uses a Canopy Height Model (CHM), an altimetric selection mask, weights, a low-pass filter, and specific algorithms to generate a DTM from a high-resolution DSM derived from the UAVs and a DTM of a 1:50,000 map. With the DTM thus obtained, landslide susceptibility assessment was then conducted. The assessment completed by means of multiple logistic regression (MLR) in the study area. The Cerro de la Miel in Tepoztlán, State of Morelos, Mexico, is selected to exemplify this method. The study area was affected by rockfalls and shallow landslides during the earthquake on September 19, 2017. The results show an adequate representation of the ground topography, and eliminating most of the noise coming from the high-resolution DSM allowed us to define the landslide susceptibility. For the calculated landslide susceptibility, there is a 76% match between the model and the landslide inventory.