Dynamic development of landslide susceptibility based on slope unit and deep neural networks

Abstract

The Three Gorges Reservoir is one of the areas with the most serious landslide hazards in China. Landslide susceptibility indicates where landslides are prone to occur in the future under the influences of certain geoenvironmental and triggering conditions and is an important way for landslide prevention. This work employs multi-source and three-temporal landslide monitoring data (geology, terrain, hydrology, and remote sensing data) to reveal the dynamic change of landslide susceptibility with time in the Badong-Zigui section in the Three Gorges area. Nine influence factors for landslides (land use, aspect, engineering rock group (ERG), slope, distance to river (DTR), relative relief, normalized difference water index (NDWI), normalized difference vegetation index (NDVI) and annual cumulative rainfall (ACR)) are generated from the monitoring data. The algorithms of slope unit segmentation and deep neural networks are adopted to conduct landslide susceptibility evaluations in the 3 years of 2002, 2007, and 2017 and to investigate the dynamic change of landslide susceptibility. Moreover, this work also reveals the dynamic response of landslide susceptibility to the changing factors of rainfall, reservoir water fluctuation, soil moisture, and land use. Some new viewpoints are suggested as follows. (1) The main factors affecting landslide occurrence are DTR, NDWI, relative relief, and ERG. Among them, DTR contributes most in all the 3 years; thus, reservoir water fluctuation has the most important impact on landslide occurrence in the study area. (2) From 2002 to 2007, the new high-susceptibility areas mainly appeared along the Yangtze River and also distributed around the roads. From 2007 to 2017, more than half of the new high-susceptibility areas were distributed around the roads, and susceptibility increases also occurred in the mountainous areas far from the Yangtze River. (3) The development of landslide susceptibility from 2002 to 2007 was mainly caused by the rising of reservoir water level as well as road construction. The change of landslide susceptibility from 2007 to 2017 was mainly caused by rainfall and road construction. This work may provide some clues on landslide prevention and control according to the dynamic development of landslide susceptibility and the causes of the susceptibility changes.