Landslides on dry badlands: UAV images to identify the drivers controlling their unexpected occurrence on vegetated hillslopes

Landslides are geomorphological processes that consist in the mobilization of ground, rocks, debris, and mud downslope that cause local erosion problems. The eroded materials can be transported downstream, which implies an additional environmental risk that might lead to catastrophic and significant economic and human losses. Rainfall is usually the main triggering factor for landslides occurrence, but there are other factors such as previous weathering activity, topography, soil properties and vegetation which also play a key role. These drivers have been deeply studied in humid regions, but little is known about the factors controlling landslide incidence in arid and semi-arid environments. The aim of this work was to identify the incidence of landslides in a semiarid badlands system in the SE of the Iberian Peninsula to identify key drivers of this process and to develop a predictive landslide susceptibility model. To do this we built a landslide inventory after an unusual rainy period (five consecutive rainy days representing about 50% of mean annual rainfall in the area). Then, a photogrammetric UAV-based flight allowed us to estimate the total extent of identified landslides, as well as, vegetation coverage distribution and several topographic attributes that were known to influence landslide formation. Our results prove that vegetation cover plays a key role in the occurrence of landslides during prolonged periods of consecutive large amount and low intensity rains. Vegetation in the area is mainly dominated by small shrubs and grasses that prevent water erosion and infiltrated run-on water from upstream areas during most rainfalls. However, most dominant species are characterized by shallow root systems that tramp and increase soil water storage mainly in the upper soil layer but do not retain soil further than surface layers. During prolonged very high magnitude and low intensity rainfalls, such as these studied here, water accumulated in the first 30–50 cm of soil until saturation occurs, favor landslide formation. Topographic variables such as elevation, slope gradient and aspect are also of paramount importance controlling landslide incidence, with most landslides occurring in intermediate parts of steep N and N–W hillslopes. Finally, by combining topographical attributes and the vegetation map obtained from the low cost UAV flight it was possible to developed a non-parametric multivariable model that accurately predicted landslide occurrence during prolonged periods of low intensity rainfall by providing a landslide susceptibility map. This low cost method can be extrapolated to other environments with similar characteristics to foresee these processes. A prevention method as the one developed in this work is key for a correct management with areas with high risk of occurrence of landslides.