Refinement analysis of landslide risk assessment for wide area based on UAV-acquired high spatial resolution images

The Loess Plateau is the largest loess accumulation zone globally. It has a fragile geological and ecological environment, experiences significant water and soil loss, and is prone to frequent landslides and collapses. Thus, landslide risk assessment and disaster prevention and reduction are required in this region. Using images acquired from unmanned aerial vehicles (UAVs) has the advantages of low cost, flexible data collection, high spatial image resolution, and real-time image data over traditional landslide risk assessment methods. UAV remote sensing has been used to identify and extract single or small loess landslides and determine elements at risk. An effective method is required to conduct wide-area landslide research for land-use planning. We used high spatial resolution (0.13 m) UAV images and Geographic Information Systems (GIS) analysis to update landslide catalog data and extract land use, roads, rivers, and other elements at risk. The frequency ratio coupled with the random forest model was used to evaluate landslide susceptibility; the prediction accuracy was high. The area under the curve (AUC) was 0.791. The risk index was calculated for five rainfall intensities, and the vulnerability evaluation and value estimation of the element at risk were completed by grey correlation model. Susceptibility, hazard, and the loess landslide vulnerability evaluation and value estimation of the elements at risk are combined to realize the fine evaluation of the whole process of the wide-area (164 km²). This study demonstrates that combining high spatial resolution UAV images and GIS is suitable for wide-area loess landslide risk assessment. This approach can be used for wide-area refined risk assessment of loess landslides in areas with similar geological conditions.