Regional landslide hazard assessment often lacks short-term responsiveness because susceptibility products describe long-term predisposition, whereas recent deformation fields are delayed by SAR acquisition and processing. This study develops a dual-scale workflow for regional landslide hazard that links a long-term hazard baseline with short-term deformation updating. Landslide susceptibility and time-series InSAR define the long-term hazard background, while a spatiotemporal deep learning model predicts deformation increments from historical deformation and dynamic triggers. Ablation analysis identifies deformation history combined with dynamic triggers as the most effective short-horizon configuration. A two-level matrix then updates the long-term baseline using the predicted deformation increments. The resulting workflow preserves the long-term deformation background while providing temporally responsive hazard updates for slow-moving landslides at the regional scale, supporting regional hazard identification and risk management in complex terrain.
Landslides are among the most destructive geohazards in open-pit mines, and their monitoring and early warning are of great importance for risk mitigation. Video surveillance systems integrated with machine vision technologies offer innovative solutions for monitoring the stability of rock slopes. In this study, a dataset of 36 documented rockslides was compiled, revealing that rockfall activity occurred prior to failure in all cases, confirming its role as a consistent and reliable precursor. To detect rockfalls from surveillance videos, a low-computational method combining sparse optical flow and fast background subtraction was developed. Compared to traditional motion detection methods, this rockfall tracking algorithm is designed especially to compensate for challenging conditions such as camera vibration and extreme weather disturbances. The short-term average/long-term average (STA/LTA) method was employed to detect and amplify variations in rockfall frequency, enabling automatic early warning. Field validation was conducted through two representative case studies including a bench slope rockslide at the Zhongshan Quarry and an inter-ramp slope rockslide at Yanshan Iron Mine. The results indicated that rockfall monitoring can provide timely landslide early warnings, with lead times ranging from seconds to hours, exceeding the minimum evacuation time required. Specifically, lead times of 21 s and 91 s were determined for sliding body heights of 6 m and 30 m, respectively. The proposed video-based landslide early warning system demonstrated high application potential in open-pit mines.
Granite residual-colluvial soil in the northern piedmont of the Qinling Mountains forms stratified mantles prone to shallow landslides. Field observations showed failures at different stratigraphic positions, but the controls remain unclear. Investigation of 18 profiles identified three configurations: A–B–C–R in eight profiles, B–C–R in five, and C–R in five; A, B, and C denote the humic, illuviated, and parent-material horizons, and R denotes weathered bedrock. One A–B–C–R profile was selected for comparison of grain size, mineral composition, CT-resolved pore structure, hydraulic behavior, and shear strength. The A horizon had the highest CT-resolved porosity, lowest tortuosity, and most continuous pore network, favoring distributed flow near saturation. The B horizon contained the finest dominant fraction, highest clay-mineral content, and stronger pore-throat constriction, while showing higher model-estimated conductivity than the A horizon at intermediate suction. The C horizon had the lowest CT-resolved porosity and pore-network connectivity, although planar fissures produced channelized flow. Saturated hydraulic conductivities were 1.80 × 10–3, 1.02 × 10–3, and 6.68 × 10−4 cm s−1 for A, B, and C, respectively; A was 1.76 times B and 2.69 times C. Unsaturated conductivity varied with suction. These contrasts indicate hydraulic transitions at A–B and B–C contacts and provide conditions for water redistribution and suction loss. At natural water contents, apparent cohesion and apparent friction angle increased from A to C. The integrated evidence supports two failure-localization modes: interface-associated sliding along stratigraphic contacts and intra-horizon shearing. Failure localization is associated with the preserved stratigraphic configuration and the corresponding hydraulic and mechanical contrasts. These findings provide a process-based interpretation of potential shallow-landslide failure localization in slopes with comparable granite residual-colluvial soil stratigraphy.
On 4 July 2023, extreme short-duration rainfall triggered a bedding rock landslide in Changtan Town, Wanzhou District, Three Gorges Reservoir Area, Southwest China. Rainfall reached 261.2 mm in 24 h, with a peak intensity of 57.6 mm/h. The landslide mobilized approximately 3.03 × 105 m3 of material, destroyed seven buildings, buried about 200 m of road, and caused one fatality. Field investigation, aerial image interpretation, DEM analysis, E-SBAS InSAR, and DAN3D simulation were integrated to examine the deformation evidence, failure mechanism, runout, and residual hazard. Failure was controlled by dip-slope structure, an argillized shale layer, contrasting permeability between highly and moderately weathered shale, and a pre-existing free face at the shear outlet. No reliable spatially coherent precursor was detected at the available E-SBAS measurement points during the available pre-failure observation period; localized deformation outside these points cannot be excluded. DAN3D back analysis reproduced the principal runout and depositional pattern of Zone A, with a maximum velocity of 11.91 m/s and an area overlap ratio of 0.75. Assumed failure scenarios for potential sliding zones B and C produced a maximum velocity of 9.49 m/s and a maximum depositional thickness of 5.58 m. Their combined residual hazard zone intersects eight buildings and several road sections. The integrated analysis supports scenario-based residual hazard assessment for rapid rainfall-induced bedding rock landslides.
A large retrogressive landslide occurred in sensitive glaciolacustrine clay deposits along the Blanche River, 7 km north of Englehart, Ontario, Canada, early on June 20, 2025. The failure took place on the outer bank of a tight river bend and had an estimated volume of between 380,000 and 475,000 m3. The landslide is classified as an extremely rapid retrogressive flow slide, typical of larger sensitive clay landslides elsewhere in eastern Canada. The depletion zone was almost fully evacuated, exposing near-horizontal sliding surfaces, with some intact rafts of trees on thin soil layers. Landslide debris flowed both upstream and downstream over several hundred metres and dammed the Blanche River. The dam was overtopped after about 34 h, with gradual incision occurring over the following weeks. The specific trigger of the landslide was not identified. Occurring within a group of 16 retrogressive landslides along 14 km of river length of the Blanche and lower Englehart rivers, this landslide confirms that rapid retrogressive failures in glaciolacustrine deposits are active processes within the glacial Lake Barlow clay belt, northeastern Ontario. The June 2025 landslide warrants further study and investigation. An improved understanding of recent landslides can lead to effective management of future landslide risks.
