Single-phase-to-ground (SPG) faults are the most frequent fault type in medium-voltage small-current grounding distribution networks. Due to arc suppression coil compensation and nonlinear grounding behaviors, fault currents are weak and highly nonstationary, making accurate fault type identification challenging under complex operating conditions. This paper proposes a physics-inspired multimodal collaborative representation framework based on the Hilbert–Huang Transform (HHT) for SPG fault type identification. A typical 10 kV distribution network model is established in PSCAD/EMTDC to generate multi-scenario fault samples. HHT is applied to construct complementary dual-modal features, including time–frequency energy maps and instantaneous temporal evolution sequences. A parallel Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture is developed to model spectral–spatial patterns and temporal dependencies, while channel and temporal attention mechanisms are introduced for adaptive feature enhancement. Results show that the proposed method achieves an overall accuracy of 99.2% across five fault types. Under additive white Gaussian noise with signal-to-noise ratios down to 5 dB, the accuracy remains above 93.6%. These simulation-based results indicate that the proposed framework maintains robust performance within the modeled PSCAD/EMTDC scenarios.

Discontinuity trace is a critical geological information element in engineering geology. However, existing point cloud-based methods still face challenges, including insufficient trace connectivity and ambiguous topological relationships between trace. To address these limitations, this paper proposes a voxel-based morphology-topology approach instead of complex point connection algorithms. Initially, we coarsely extract trace points by applying the elbow method to optimize K-means++ clustering, which based on normal vector neighborhood deviations of point clouds. After voxelizing the trace points, we enhance and refine spatial connectivity through morphological dilation and an improved secondary erosion process. We then define Euler characteristics to characterize the topological properties of the voxels. Finally, obtain a simplified voxel skeleton containing trace points through topological iterative deletion, generate vectorized trace paths by connecting voxel centers, followed by post-processing. We validate the method through three slope cases, the topological connectivity rate compared to existing methods is the highest reaching 90.48 %, 90.91 %, and 91.97 %, respectively. demonstrating its ability to identify finer trace details more precisely than alternative approaches. The proposed method enables robust extraction of vectorized paths from the trace spatial skeleton, which shifting from connecting points to simplifying a topological structure. This approach offers reliable technical support for digital geological logging and intelligent assessment of engineering stability.

Reverse-fault rupture is commonly associated with large-scale landsliding. Yet, the dynamic hillslope response to seismic shaking during reverse fault dislocation remains elusive, leaving our understanding of one of the most destructive landslide triggers incomplete. Here, we address this gap with a physical experiment that simulates the coseismic response of hillslopes to an intersecting reverse-fault rupture. We employ a bespoke shaking-table apparatus with a sliding bottom plate to simulate reverse fault dislocation, utilizing high-speed cameras and high-definition particle image velocimetry for monitoring. Our results show that, owing to reverse fault rupture, the acceleration amplification is 1.2–2.6-fold greater on the hanging wall relative to the footwall in the dynamic response of the hillslope. This amplification leads to the development of tensile cracks in the hillslope, which initiates large-scale landsliding and the sudden projectile-like ejection of mass via the trampoline effect.

Landslides induced by long-term agricultural irrigation are widespread on the Heifangtai loess terrace in Gansu Province, northwestern China. Early identification of active and potentially reactivated loess landslides, together with an explanation of their deformation mechanisms, is essential for hazard assessment and mitigation. This study integrates high-resolution optical interpretation, small baseline subset interferometric synthetic aperture radar (SBAS-InSAR), and unmanned aerial vehicle (UAV)-derived geomorphic evidence to investigate the spatial distribution, deformation evolution, and failure modes of representative irrigation-induced loess landslides. Optical images were used to delineate landslide boundaries, rear scarps, tension cracks, damp zones, and toe fissures, whereas 192 Sentinel-1A ascending scenes from October 2014 to June 2023 were processed to derive long-term deformation fields and time-series displacement curves. Three representative landslides, DC#2, JJ#4, and CJ#8, were selected because they are located in different active landslide clusters, have documented failure histories, and show clear optical, UAV, and InSAR evidence of deformation. The results indicate that: (i) diagnostic geomorphic features identified from optical imagery provide effective spatial constraints for interpreting InSAR point targets; (ii) representative active landslides show marked spatiotemporal heterogeneity, with line of sight (LOS) deformation velocities generally ranging from approximately − 15 to − 50 mm/yr, indicating slow-moving deformation or active creep rather than instantaneous rapid motion captured by InSAR; and (iii) the dominant deformation mechanism is the rise in groundwater levels caused by long-term irrigation, while freeze–thaw cycles and rainfall infiltration act as important short-term accelerators. Cross-validation using optical interpretation, UAV orthophotos, historical landslide records, and published GPS/GNSS, crack gauge, and field evidence supports the general reliability of the detected deformation patterns. Although single-viewing-geometry InSAR cannot fully resolve three-dimensional displacement, its combination with optical and UAV-derived evidence provides a practical framework for regional-scale identification and monitoring of irrigation-affected loess landslides.

Loess is the primary material used for engineering construction on the Chinese Loess Plateau. To address environmental issues posed by solid waste and the poor engineering properties of loess, solid waste materials from power plants were utilized to enhance loess, thereby achieving comprehensive resource utilization. In this study, the utilization of solid waste materials to improve loess is examined, and their properties, microstructure, and stabilization mechanism at different maintenance times are investigated. First, orthogonal tests were conducted on the unconfined compressive strength (UCS) of loess specimens containing cement, fly ash, and solid waste material A to derive the optimal composite proportioning scheme. Subsequently, a series of mechanical, hydraulic, and durability tests was conducted on the cured loess to evaluate the effect of the composites on the loess. The results showed that fly ash-based composites effectively enhanced the mechanical properties of loess during compression, bestowing the loess with enhanced impermeability and durability. The UCS of the improved loess specimens reached 8.25 MPa after 28 d of maintenance, while the permeability coefficient decreased to 1.01 × 10-8 cm/s. The strength loss index of the specimens was 95.02% after 96 freeze-thaw cycles. Finally, microscopy tests were conducted to reveal the physical and chemical mechanisms by which the composites improved the loess. The composites produced numerous new gels that fill the internal pores of the loess and strengthen the connections between soil particles, resulting in a denser and more stable internal structure. The proposed methodology utilizes solid waste materials from power plants, such as fly ash, as amendment materials for secondary use and in situ abatement of loess, thereby offering environmental and economic benefits.

On 8 February 2025, a catastrophic landslide occurred in Junlian County, Sichuan Province, southwestern China, resulting in 10 fatalities, 19 missing persons, and severe damage to houses and farmland. This event offers a rare opportunity to investigate the full dynamics of a high-speed, long-runout landslide from direct observations. Previous studies relied mainly on geological and remote-sensing analyses of pre- and post-failure data, but the lack of direct observations during failure has limited understanding of its dynamic evolution. In this study, we integrate broadband seismic records capturing the complete failure process from the Sichuan regional network with high-resolution airborne LiDAR data to reconstruct the complete dynamics of the Junlian landslide. The seismic signals reveal two major failure episodes (LS01 and LS02), initiated at 11:50:54.855 and 11:53:38.615 (UTC+8), respectively. Detailed analysis of LS01 identifies four kinematic stages and two discrete sliding masses, whose volume ratio ranges from 0.296 to 0.304, as validated by spectral and 3D geometric analyses. These results provide a comprehensive reconstruction of the temporal evolution of the landslide and demonstrate the potential of seismic observations to quantitatively resolve the dynamics of catastrophic landslides in near-real time.

The construction of an underground utility tunnel system has been proposed for Xi’an, a major transportation center in western China. However, this city is situated in the Fenwei Basin and fault activities dominated by the Lintong-Chang'an Fault directly threaten underground engineering safety. The deformation of prefabricated utility tunnels under seismic action is dominated by transverse shear and the haunch angle is more prone to failure, while the damage evolution law remains unclear. To investigate the seismic performance of utility tunnels, quasi-static graded loading and constant amplitude loading model tests were established to study the stress and deformation characters of prefabricated underground utility tunnels under transverse seismic excitation. Analysis of concrete cracking, reinforcement stress, and structural deformation revealed that the stiffness of the utility tunnel degrades more rapidly under graded loading, corner plastic rotation and plastic cracking of the sidewall dominating the deformation, and crack propagation may cause local instability whereas basalt fiber can effectively suppress micro-crack propagation and improve the overall stability of the tunnel.

The characteristics of coplanar discontinuities in rock masses are critical for understanding the stability of underground excavations. Digital window mapping provides a rapid, intuitive representation of rock mass discontinuities and their influence on failure kinematics. In this paper, we demonstrate a novel method for structural rock mass mapping using 3D point clouds obtained from terrestrial laser scanning (TLS) at an underground rock outcrop in China. We introduce an improved watershed clustering algorithm for automatic identification of rock mass discontinuities, with a workflow to analyze coplanar characteristics of rock mass discontinuities and perform rapid window mapping, called the Progressive Local Coplanar Search for Discontinuities (PLCSD) method. The method can identify one or more sets of rock mass discontinuities that are approximately coplanar in 3D space to solve the problem of over-segmentation of discontinuities. Through spatial projection, coordinate transformation, and coplanar trace linking, the discontinuities of the rock outcrop are automatically mapped onto a window canvas. The results indicate that the median dihedral angle difference between automatically identified individual rock discontinuity surface and manually measured mapping ranges from 4.06 degrees to 7.25 degrees; The average optimal classification accuracy of the coplanar analysis is 0.924. However, the analytical results are not strictly optimal, as they depend on the combined of two key parameters: the angle delta between discontinuity normal vectors and the average discontinuity spacing L. The digital window mapping method effectively captures rock mass discontinuities and small-scale folds, supports further vectorized mapping with other discontinuity identification methods, and offers a high-efficiency alternative to traditional mapping techniques.

During 26–28 July 2024, an exceptional rainstorm associated with Typhoon Gaemi triggered widespread clustered landslides in five eastern towns of Zixing County, Hunan Province. However, quantitative analysis of the controlling factors behind such extreme rainstorm-triggered clustered landslides remains insufficient. In this study, a detailed landslide polygon inventory was effectively established using high-resolution satellite imagery and the deep learning algorithm, which was subsequently validated by field investigations. An interpretable machine learning framework, integrating XGBoost algorithm with SHAP (SHapley Additive exPlanations), was developed to quantify the relative contributions of eight critical controlling factors. Spatial analysis reveals that landslides predominantly occurred at elevations of 400–800 m, slope gradients of 30–40°, and south-to-southwest-facing slopes. Furthermore, they were closely associated with Caledonian granite strata near fault zones and areas receiving over 500 mm of rainfall. The XGBoost model demonstrated robust predictive performance across five-fold cross-validation. SHAP analysis identified 3-day cumulative rainfall as the most significant controlling factor, followed by lithology, distance to faults, and slope aspect. Based on the coupling analysis of these factors, six factor combinations were identified to estimate landslide probability. This multi-method investigation advances our understanding of cascading meteorological-geological hazards and highlights critical controlling factors driving clustered landslide occurrences in subtropical mountainous regions.

Quasi-periodic slip on some natural faults is thought to originate from the stick-slip instability of granular gouge. Laboratory observations further suggest that grain-to-grain contacts generally exhibit time-dependent friction, which can be empirically described by the Rate-and-State Friction (RSF) law. To investigate how such behavior controls stick-slip, we incorporate RSF into a discrete element framework with linear elastic contacts, enabling simulation of laboratory-like stick-slip in granular systems. The results suggest that pre- and co-seismic slip are controlled by the dynamic changes of the micro-friction coefficient (& micro;micro). In a single stick-slip cycle, & micro;micro exhibits four velocity- and two state-dependent stages, inwhich & micro;micro first strengthens and then weakens with relative slip rate (vrel) during co-seismic slip, while the contact state (B) continuously weakens. Interestingly, the spatial evolution of vrel and B shows that contact-state weakening evolves synchronously with slip acceleration at the rupture front, suggesting a progressive self-weakening of co-seismic slip. Under different loading conditions, the contact state strengthens at lower loading rates and higher normal stresses, consistent with the friction response. Furthermore, the stress drop and fracture energy (EG) increase with co-seismic slip distance, with loglinear slopes of 0.94 and 1.75. The larger EG relative to smooth faults likely reflects more extensive grain-contact failure. The study provides mechanistic insight into how grain-scale friction controls fault stick-slip in granular gouge.

On 7 August 2025, an extreme rainfall event impacted Yuzhong County, Gansu Province, China, triggering cascading hazards in the Hongzhuangzi Gully, including flash floods, widespread shallow landslides, the formation of a temporary channel-blocking dam (CBD), and a subsequent debris flow. UAV photogrammetry, high-resolution satellite imagery, rainfall records, and field surveys were integrated to reconstruct the event’s four-stage evolution. A total of 120 shallow landslides were identified, mainly on 25–35° slopes with east to southeast aspects, with failure depths mostly less than 0.6 m. These landslides were concentrated in areas where thin colluvium overlies bedrock, suggesting a possible link to slope hydrological responses under extreme rainfall conditions. In the mid-gully reach, dense plantation forests and a natural constriction may have contributed to temporary blockage by facilitating the accumulation of woody debris and sediment. Geomorphic indicators including mud lines up to 4.5 m high, upstream boulder clusters, asymmetric bank erosion, and a downstream debris fan provide indirect evidence supporting the formation and subsequent failure of a temporary CBD. RAMMS simulations, calibrated against field observations, estimate a total debris-flow volume of approximately 1.105 × 10⁶ m3, with about 82

Based on core observations, thin section identification, well logging and seismic data of the Permian strata in the western Sichuan Basin, this study systematically discusses the constraints of Middle and Late Permian two-phase volcanism on paleotectonic evolution and sedimentary filling. It further reveals a long-neglected Chengdu–Santai volcanic paleo-uplift developed continuously in the study area and clarifies its controlling effects of formation and evolution on sedimentary reservoirs. The results indicate that two suites of volcanic rocks are developed in the Permian strata of western Sichuan Basin. The lower volcanic rocks of the Middle Permian Maokou Formation are predominantly composed of basalt and volcanic breccia, whereas the upper volcanic rocks of the Upper Permian Longtan Formation are mainly basalt and volcaniclastic rocks, recording two episodic volcanic eruption events that occurred in the middle depositional stage of the Maokou Formation and the early depositional stage of the Longtan Formation. Paleogeomorphological reconstruction shows that the volcanism during the middle Maokou depositional stage initially formed the embryonic paleo-uplift on the western Sichuan carbonate platform. Continuous vertical stacking and lateral expansion of volcanic materials during the early Longtan depositional stage finally shaped a sub-elliptical paleo-uplift plunging northeastward. This paleo-uplift extends approximately 200 km northeastward along its major axis, with a minor axis ranging from 70 km to 110 km in the east-west direction. The uplift high is located in the Well area YS1–YT1, with a maximum relative relief of 300 m. Affected by late tectonic activities and persistent sedimentary filling and leveling, the paleo-uplift completely disappeared during the deposition of the third member of the Feixianguan Formation. Formed by the synergistic effect of the clockwise northward rotation and drift of the South China Plate and mantle plume hotspot activity during the Permian, this paleo-uplift dominated the paleogeographic pattern of the central and western Sichuan Basin during its development. Specifically, tidal flat-barrier shoal sedimentary systems developed around the volcanic highland during the middle to late Maokou depositional stage, and high-energy facies belts were widely distributed around the volcanic paleo-uplift during the Changxing depositional stage. This study confirms that the volcanic paleo-uplift and its peripheral zones are favorable hydrocarbon exploration areas for the Maokou and Changxing formations, providing a new perspective for Permian oil and gas exploration in the Sichuan Basin.

Conventional landslide susceptibility maps for China’s Three Gorges region primarily rely on static inventories and slope-gradient analysis, potentially overlooking other contributing factors. Using seven machine-learning models with eleven conditioning factors, we found that elevation and monsoon rainfall showed the strongest statistical association with historical landslide distribution, together accounting for approximately 90

Accurate landslide detection is critical for geological hazard early warning, yet existing deep learning methods lack systematic architectural evaluation and inadequately account for morphological variations. This study proposed a multi-module synergistic architecture integrating Atrous Spatial Pyramid Pooling (ASPP), Squeeze-and-Excitation (SE) attention, and Path Aggregation Network (PANet) with an EfficientNet-B4 encoder for automated landslide segmentation from high-resolution satellite imagery. Comprehensive ablation experiments on the Bijie landslide dataset quantified individual module contributions: SE attention improved Intersection over Union (IoU) by 2.95

Earthquakes and seismically-induced landslides accelerate carbon export from mountains by eroding hillslope soil carbon. However, a quantitative understanding of their net contribution to carbon cycling remains incomplete. Using the 2008 Mw 7.9 Wenchuan Earthquake which generated the largest landslide volume in recent history, we quantify its carbon mass balance accounting for storage, loss, and transport within the ensuing sediment cascade. Thanks to post-event revegetation and extensive intermontane sediment storage, we show that the earthquake boosted Longmenshan carbon mass by 10

Microcontinents are isolated fragments of continental crust surrounded by oceanic lithosphere. They commonly occur in modern ocean and are also identified within orogenic systems. During ocean-continent subduction processes, microcontinents can be accreted onto continental margins through collision and subduction, leading to the migration of subduction zone toward the oceanic domain. However, the potential for this process to be preserved in metamorphic records remains poorly understood. This study focuses on newly identified Datong-Mengyuan terrane (DMT), a high grade metamorphic-magmatic terrane located along the northern margin of the conventionally defined Qilian block (QLB). The DMT is tectonically separated from the low- to medium-grade metamorphic basement of the QLB by a dextral strike-slip ductile shear zone and an ophiolite mélange. The petrological analysis and texturally-controlled U-Pb multi-mineral geochronology reveal that the mafic and felsic granulites from the DMT preserve evidence of two distinct metamorphic events: an early high-pressure granulite-facies event (11.4–13.7 kbar, 735–805°C) at ~500 Ma, and a later low- to medium-pressure granulite-facies overprint (5.5–9.6 kbar, 790–840°C) at ~460–450 Ma. We interpret the Cambrian high pressure granulite facies metamorphism as resulting from collisional thickening during the accretion of the DMT to the Qilian block. In contrast, the subsequent lower- pressure overprint reflects decompression heating in a continental arc setting, likely associated with the oceanward migration of the subduction zone following microcontinent accretion. These findings provide crucial insights into how metamorphic processes record both microcontinent accretion events and the dynamic evolution of convergent plate boundaries.

Earthquakes, extreme rainfall, and other conditions can trigger numerous shallow landslides, posing significant safety hazards. Non-obvious landslides, which lack clear precursory signs, are particularly challenging to predict. While traditional remote sensing and UAV aerial surveys can detect surface deformations, they fail to anticipate the occurrence of non-obvious landslides due to their inability to probe subsurface structures. Therefore, exploring slope underground structural characteristics is the key to studying non-obvious landslides. This study investigated the Zhongzhai landslide in Niangniangba, Qinzhou District, Tianshui City, Gansu Province, and the apparent resistivity of profile of the landslide area was obtained by electrical resistivity tomography survey, using in-situ light dynamic penetration test and drilling data to confirm the relationship between resistivity and formation lithology. the formation structure was then obtained from the three-dimensional model of high-density resistivity. UAV photogrammetry and field investigations provided terrain features, facilitating the reconstruction of a three-dimensional geological model of the Zhongzhai landslide, to realize the fusion of multi-source data. Numerical simulations were then conducted to explore the development process and failure mechanism of landslides at the loess-bedrock interface. The resistivity value at the interface between bedrock and loess cover in the study area is approximately 200 Ω·m. The development of the Zhongzhai landslide is closely related to the geological structure, and the spatial characteristics of the bedrock–cover interface play a decisive role in rainfall-induced landslides. The proposed multi-source data fusion method and three-dimensional modeling approach provide an effective reference for the stability evaluation and prediction of non-apparent landslides.

On 8 July 2025, an uncatalogued glacial lake (0.83 km2) in Gyirong, Tibet, burst and destroyed the China–Nepal Friendship Bridge, causing at least 9 deaths and about 30 persons missing (per official reports as of August 2025). The lake was absent from inventories not because a 1 km2 threshold excluded it, but because it had only recently formed, and because several regional inventories have incomplete coverage below roughly 10 ha and lag behind the rapid formation of supraglacial and newly formed moraine lakes between update cycles. A single-day rainfall of 194 mm on 28 September 2024 triggered persistent soil moisture anomalies and a nine-month hydrological preconditioning period. HEC-RAS-simulated inundation is qualitatively consistent with NDVI-derived vegetation damage in the main affected reach, providing independent corroboration of the downstream impact extent, while mismatches in shadowed gorges and multi-branch sections reflect DEM-resolution limits. This case suggests that: (1) where sub-threshold lakes coincide with debris supply, channel confinement, and exposed infrastructure, downstream impacts can be disproportionate to lake area (based on this event together with documented analogues such as Chorabari and Gongbatongsha); (2) a roughly six-month deformation precursor was identifiable in this case, although its transferability to other moraine-dammed systems requires testing across larger samples; and (3) dynamic monitoring of inventory-gap lakes and transboundary data-sharing mechanisms along strategic corridors are urgently needed.