From a global perspective, the basic mapping and investigation of the loess sinkholes are far less extensive and in-depth than those of karst sinkholes. To some extent, this has limited people's understanding of the morphological characteristics, development patterns, and formation mechanisms of the loess sinkholes. The Chinese Loess Plateau (CLP) features the most typical loess landforms in the world, where tens of thousands of loess sinkholes have developed. However, due to the lack of high-precision and high-resolution survey data, the identification, characterization, and quantification of sinkholes in the CLP are basically blank, which significantly hinders in-depth research on loess sinkholes. In this study, we investigated a typical watershed in the CLP using photogrammetry, airborne laser scanning, and a handheld laser scanner. Based on previous studies, this paper introduces indices and methods for the morphological quantification of loess sinkholes and constructs the first-ever dataset of loess sinkhole morphology containing 1194 records at the basin scale. On this basis, we completed the spatial mapping of loess sinkholes, analysis of distribution patterns, morphological analysis, size-frequency analysis, fitting analysis of different parameters, estimation of subsurface soil erosion, in-depth investigation of typical sinkholes, and quantification of the contributions of different factors to sinkhole development. These efforts provide rich information for a deeper understanding of the morphological characteristics and genesis of loess sinkholes and offer data support for comparative studies with sinkholes in other regions. More importantly, we preliminarily estimate that the subsurface soil erosion triggered by sinkholes in the study area reaches as high as 345 000 metric tons. This finding underscores that loess sinkholes are not only a geological disaster but also a serious form of soil loss, highlighting their undeniable significance in regional soil erosion studies and laying a solid foundation for subsequent research and disaster prevention efforts. Furthermore, we suggest that the integration of airborne laser scanning and handheld laser scanning may represent a new trend in the detailed investigation of sinkholes in the future. This dataset is available on the Zenodo platform (10.5281/zenodo.14000267, Hu et al., 2025).

The Loess Plateau is a major engineering intensive and energy producing region in China, where landslide risk has become increasingly prominent. Therefore, landslide susceptibility assessment is of great importance for ensuring regional safety and sustainable development. However, most existing landslide susceptibility models focus on improving prediction accuracy while neglecting the spatial heterogeneity of landslide occurrence. In this study, Yaozhou District, China, was selected as the case study. A dataset comprising 16 conditioning factors related to topography, geology, land use, and vegetation coverage was constructed. Two spatial heterogeneity aware models, namely geographically weighted random forest (GWRF) and geographically weighted categorical boosting (GW-CatBoost), were developed based on geographically weighted regression. These two models were compared with traditional global machine learning models, namely random forest (RF) and CatBoost, in terms of landslide susceptibility prediction performance. In addition, SHapley Additive exPlanations (SHAP) analysis was used to interpret the contributions of conditioning factors in each model.The results showed that vegetation coverage, human activity and road related factors were the dominant conditioning factors of landslides in the study area, and that the geographically weighted models further revealed distinct spatial variations in the dominant conditioning factors and local landslide controlling mechanisms. A unified evaluation framework including AUC, confusion matrix, and derived metrics was adopted. GW-CatBoost achieved the highest AUC point estimate of 0.926 and performed favorably across most confusion-matrix-derived metrics. The performance of both geographically weighted models remained relatively stable when the number of retained conditioning factors was reduced from 13 to 8, while GW-CatBoost maintained an AUC of approximately 0.87 with 5 dominant factors. The proposed framework addresses a key limitation of global machine learning approaches, which assume constant effects of conditioning factors across space, and reveals spatially heterogeneous patterns of factor sensitivity across subregions. Its flexible factor configuration and regionally adaptive structure also provide potential for application in other geographical contexts. These findings provide a scientific basis for refined hazard management and regional development planning in the Loess Plateau area.

Abstract. From the perspective of the world, the basic mapping and investigation of the loess sinkhole is far less extensive and in-depth than that of the karst sinkhole survey. To some extent, this hinders people’s understanding of the morphological characteristics, development rules, and formation mechanisms of the loess sinkholes. Chinese Loess Plateau (CLP) has the most typical loess landform in the world, and tens of thousands of loess sinkholes have developed. However, due to the lack of high-precision and high-resolution survey data, the identification, characterization, and quantification of sinkholes in the Loess Plateau are basically blank, which seriously hinders the in-depth study of loess sinkholes. We investigated a typical watershed on the Chinese Loess Plateau using photogrammetry, airborne laser scanning, and handheld laser scanner. Based on previous studies, this paper proposes indices and methods for the morphological quantification of loess sinkholes and constructs the first dataset of loess sinkhole morphology containing 1194 records at the basin scale. On this basis, we completed the spatial mapping of loess sinkholes, analysis of distribution patterns, morphological analysis, size-frequency analysis, fitting analysis of different parameters, estimation of subsurface soil erosion, in-depth investigation of typical sinkholes, and quantification of the contributions of different factors to sinkhole development. These efforts provide rich information for a deeper understanding of the morphological characteristics and causes of loess sinkholes and offer data support for comparative studies with sinkholes in other regions. More critically, we preliminarily assessed that the subsurface soil erosion triggered by sinkholes in the study area amounts to as high as 345,000 metric tons. This finding makes it increasingly clear that loess sinkholes are not only a geological disaster process but also a serious soil loss process, highlighting their undeniable significance in regional soil erosion studies and laying a solid foundation for subsequent research and disaster prevention efforts. Moreover, we believe that the integration of airborne laser scanning and handheld laser scanning may represent a new trend in the detailed investigation of sinkholes in the future. The dataset is available from Zenodo platform (https://doi.org/10.5281/zenodo.14000267).

Loess is generally unsuitable as a landfill cover material because of its loose structure, high porosity, and poor cohesion. This study investigated whether a waste water-based drilling geopolymer (WWDG) can reduce loess gas permeability (GP) by hardening the soil. Triaxial GP tests were performed to quantify the effects of WWDG content (ωa), confining pressure (p), water content (ω), and dry density (ρd) on GP. The findings inform strategies for waste reutilization and loess improvement. Microstructural and compositional changes in WWDG-improved loess were characterized by scanning electron microscopy (SEM), X-ray diffraction (XRD), and mercury intrusion porosimetry (MIP). The study further clarified the curing mechanism of the WWDG-improveed loess. Results showed that the distribution of GP coefficients narrowed under the combined influence of the tested factors. ωa had the greatest impact on GP. Increasing ωa, p, ω, and ρd decreased the GP coefficients by 93.15

Due to global warming, the Qinba Mountain Area in China has experienced frequent extreme heavy rainfall events, transforming isolated geological hazards into multi-hazard chains characterized by cascading and compound disasters. These occurrences often lead to severe casualties and socioeconomic losses. To investigate the formation mechanisms of typical landslide-debris flow disaster chains in this region, this study selected the Gaojiawan disaster chain—a representative case within the study area—for systematic analysis. Using integrated methodologies, including unmanned aerial vehicle (UAV) mapping, geographic information system (GIS) spatial analysis, and a numerical simulation method based on PFC3D, the entire disaster chain process was reconstructed and investigated. The results show that collision interactions between cascading hazards were found to amplify the affected area and debris flow interactions fundamentally altered deposition mechanisms, transforming localized slope accumulation into integrated mass transport processes. This paper analyzes the transformation of the landslide debris flow disaster chain from the perspective of energy. Additionally, based on the analysis of field investigation and numerical simulation results, the formation mechanism of the Gaojiawan landslide-debris flow disaster chain is summarized into five stages: landslide formation stage, landslide activation stage, slope deposit mixing stage, debris flow formation stage, and low-speed deposition stage. This research elucidates the evolutionary dynamics and underlying mechanisms of landslide-dominated geohazard chains in mountainous regions. The proposed integrated monitoring-modeling framework provides methodological references for theoretical research and risk mitigation of similar disaster chains worldwide.

In the Loess Plateau of China, loess creep characteristics with time effects are crucial factors influencing slope stability. Especially under extreme climates, after being subjected to repeated dry-wet cycles (DWC), the creep deformation of loess is more intense, leading to large-scale and frequent disasters such as landslides. In this study, a series of direct shear creep tests was performed on the loess samples after dry-wet cycles to understand the effect of DWC on loess creep characteristics profoundly. The results show that the loess is more vulnerable to creep failure after being subjected to dry-wet cycles; the greater the normal stress of the loess, the more pronounced the effect of dry-wet cycles, and the more significant the creep deformation under shear stress. Moreover, the long-term shear strength of loess decreases exponentially as the DWC increases. Based on the Burgers model, a new nonlinear viscoelastic-plastic rheological model (NVPM model) is proposed, which can accurately describe the entire loess creep curve under the effect of dry-wet cycles, especially for the accelerated creep stage that the Burgers model cannot fit. Based on the shear creep test results, the creep parameters obtained from the NVPM model were applied to a numerical simulation of a typical loess slope. The results reveal that the NVPM model is capable of reasonably explaining the stability deformation law of a loess slope under the coupling effect of DWC and creep, and the causes of loess landslides. The research results can provide a theoretical reference for the long-term stability analysis and prediction of loess slopes.

Flow-like events significant threats to ecosystems and human life and property because of their suddenness, high speed and long distances. To study the continuous reactivation mechanism of redeposited loess, we conducted a series of rheological tests, and the results showed that the three-interval thixotropy test (3ITT) effectively captures the evolution of rheological behavior before, during, and after solid–liquid–solid phase transition. Higher moisture content was found to prolong the recovery time and enhance thixotropic strength, and recovery time for Baoji loess increased from 114 s at water content 32

Sinkholes induced by soil piping are widely distributed globally, posing significant threats to ecological environments, agricultural production, and infrastructure. Compared with karst sinkholes, soil-piping-induced sinkholes are generally smaller and less detectable, posing persistent challenges to their accurate detection and mapping using deep learning approaches. To address this issue, this study proposes a novel Multimodal Attention Fusion U-Net (MAF-UNet) that adaptively integrates spectral texture information and topographic morphological features through a cross-modal attention fusion module (CFA), enabling automatic identification, segmentation, parameter extraction, and cataloging of loess sinkholes. The proposed network adopts a dualbranch encoder architecture incorporating residual double convolution (RDC), the Convolutional Block Attention Module (CBAM), and gated skip connections to enhance feature representation. A composite loss function combining Binary Cross-Entropy (BCE), Focal, and Tversky losses is employed to mitigate sample imbalance. Experiments conducted in the Heyang loess tableland, China, demonstrate that the combination of Digital Orthophoto Map (DOM), closed depressions, and slope as input factors achieves the best performance, with a Precision of 0.92, an F1-score of 0.91, an Intersection over Union (IoU) of 0.83, a mean Intersection over Union (mIoU) of 0.91, and a Precision-Recall Area Under Curve (PRAUC) of 0.94, significantly outperforming singleand dual-factor combinations. Ablation experiments further confirm that the CFA module is the most critical contributor to model performance. Transferability experiments conducted in Laozi Gully (Huining County) and Banyan Gully (Huzhu County) demonstrate that MAF-UNet exhibits strong robustness under the complex terrain conditions of the Loess Plateau. Overall, the proposed MAF-UNet supports an end-to-end workflow for automatic sinkhole mapping encompassing data preprocessing, model training and validation, prediction, segmentation, parameter extraction, cataloging, and mapping. It provides an effective methodological framework for automatic sinkhole hazard detection and mapping in the Loess Plateau and analogous regions.

Geological hazards on the Southern Chinese Loess Plateau (SCLP), particularly loess-mudstone landslides (LMLs), are increasingly triggered by extreme climatic events. This study investigates the climatic controls on the stability of the Doujitai landslide within the active Weihe Fault Zone, a representative LML in the SCLP. A series of ring shear tests were conducted on remolded loess from the landslide’s slip zone to quantify the degradation of shear strength parameters under varying water content, normal stress, shear rate, and shear modes. The results revealed that the remolded peak and residual shear strengths first increased and then decreased with increasing water content, accompanied by a transformation from strain hardening to strain softening. The remolded peak strength increased significantly with shear rate, while residual strength first decreased and then increased. Increasing shear rate also enlarged the amplitude of residual shear stress fluctuation. Compared with single-stage tests at different normal stress levels, multi-stage shear led to an attenuation of internal friction angle. Based on the experimental data, a hydro-mechanically coupled numerical analysis was conducted to model the slope’s stability evolution during rainfall. The results show an obvious relationship between the dip angle of the mudstone-loess interface and slope stability. Furthermore, the impact of rainfall intensity and duration on stability was evaluated, highlighting that higher intensity rainfall drastically accelerates stability degradation. This study elucidates the formation mechanism of LMLs under climatic control, emphasizing that extreme rainfall acts as the critical external trigger of slope instability within a tectonically preconditioned geological setting.

Rock masses are susceptible to coupled effects of acid-alkali corrosion and freeze-thaw cycles, which may degrade the internal rock structure and further threaten the long-term stability of engineering structures. This study primarily investigates the coupled effects of acid-alkali corrosion and the freeze-thaw cycle on the disintegration characteristics of red sandstone through freeze-thaw cycle tests in solutions with varying pH values, thereby revealing its micro-mechanism. This study revealed that under acidic environments, dissolution resulted in a 39.63% mass loss in red sandstone, elevated porosity to 19.65%, and induced crack formation. Alkaline conditions improved durability, resulting in merely 32.21% mass loss. Due to self-healing, porosity initially increased to 13.30% and then decreased to 4.47%. The pH value does not change significantly in neutral to weakly alkaline environments, but exhibits significant fluctuations in strongly acidic or alkaline environments. Acidic environments facilitated the dissolution of Ca2+ and Mg2+, whereas alkaline conditions enhanced the release of Si4+. The ion emission phenomenon was found to be obvious during the initial freeze-thaw stage but diminished in the subsequent stage due to mineral degradation. A hybrid grey wolf optimizer-random forest (GWO-RF) prediction model, which accounts for acid/alkali corrosion and freeze-thaw cycles, identified the fundamental mechanisms affecting disintegration behaviour and found that freeze-thaw cycles have a more significant influence than pH levels. Acidic environments inflict the most significant damage by markedly enhancing porosity, whereas alkaline environments produce gels that partially fill pores, alleviating damage. These findings provide novel experimental support for rock engineering construction and disaster prevention.

Mining-induced ground fissures in the loess regions of China have become increasingly severe under intensive coal exploitation, posing major challenges to ecological restoration. This study proposes a crack repair method that combines microbially induced calcium carbonate precipitation (MICP) with granular backfill materials (aeolian sand and loess). Triaxial shear tests and quantitative microstructural analyses were conducted on specimens repaired using different sand–loess ratios (SLS) and bacterial–cementation solution ratios (BCS). The results show that (1) Repair effectiveness is reflected in both shear failure characteristics and mechanical properties, with these differences diminishing as confining pressure increases. (2) All repaired specimens exhibited significant strength enhancement, and the maximum improvement occurred when both BCS and SLS were 1, yielding a 137 % increase in shear strength at 100 kPa confining pressure. (3) Under low confining pressure, shear strength was negatively correlated with the porosity of the repaired zone, with a minimum porosity of 12.8 % at BCS = SLS = 1. (4) BCS mainly controlled MICP mineralization efficiency and CaCO_3 crystal size, whereas SLS governed particle-size gradation; together, they regulated the pore structure and repair performance. This study provides a promising approach for the remediation of mining-induced ground fissures in loess areas.

The Yellow River “Ji-shaped Bend” region, a distinctive energy-rich zone in northern mid-latitudes, holds unique scientific significance for global studies of geological resources and loess-related hazards. Characterized by complex geology and fragile ecology, this area now experiences frequent coal-mining-induced geohazards that form loess disaster chains (LDC), that severely threatens local safety, ecology, and sustainable development. In this study, the loess disaster chain at Anshan Coal Mine was examined, analyzing its topographical features, development patterns, and disaster mechanisms. It investigates how the physical and mechanical properties of loess change under varying water contents and compaction levels. Additionally, this study explores the key factors and dynamic processes that drive the disaster chain and reveals its underlying disaster mechanisms. Findings show that after saturation, loess samples experience notable reductions in shear stress, cohesion, and internal friction angle, with cohesion decreasing the most. Comparative experiments demonstrate that lower compaction result in higher permeability and faster disintegration. The primary factors influencing the disaster chain include loess properties, extreme rainfall, geological structure, coal mining, and improper excavation, with extreme rainfall identified as the main trigger at Anshan Coal Mine.Based on these findings, the disaster chain evolution at Anshan Coal Mine progresses through five stages: 1) loess fissure formation, 2) loess landslide occurrence, 3) landslide transformation into debris flow, 4) debris flow erosion and expansion, and 5) debris flow accumulation. This study provides scientific insights for sustainable energy development and disaster prevention, while offering valuable references for disaster chain research in similar geological settings.

Loess sinkholes are extensively developed across the Loess Plateau in north-central China and have frequently evolved into chain geological hazards under intense rainfall in recent years, severely impacting regional human safety and economic development. Consequently, this study selected a typical loess sinkhole-landslide-mudflow chain (LSLMC) as the research focus. Field investigations, unmanned aerial vehicle mapping, laboratory experiments, and theoretical analyses were conducted. The scales, morphologies, types, combinational patterns, and spatial correlations of sinkholes were examined. Morphogenetic characteristics of sinkholes, landslides, and mudflows were analyzed. The hazard-inducing effects of sinkholes were revealed, and an evolutionary model for the typical LSLMC was proposed. The results indicate that: (i) The sinkholes exhibit significant spatial autocorrelation in both formation size and developmental location. Their frequency decreases as size increases. Distinct clusters of low values are observed in morphometric characteristics, whereas pronounced clusters of high values occur in elevation. (ii) The development zones of sinkholes strongly correlate with areas exhibiting a high Topographic Wetness Index (TWI) and a low Stream Power Index (SPI). The sinkholes primarily serve as preferential flow conduits and sediment transport pathways, and the interconnection of sinkholes can trigger landslides. (iii) During the formation of LSLMC, mudflow materials originate not only from loose deposits created by sinkholes and landslides but are also amplified by erosion-induced enlargement effects. The formation process of the LSLMC is summarized into six stages. This study presents the first integrated UAV, laboratory tests, and spatial correlation analysis to provide process-based insights into the formation and evolution mechanisms of the LSLMC. These findings improve the understanding of sinkhole hazard mechanisms and offer theoretical support for preventing and mitigating loess-related hazard chains in small watersheds.

Loess-red bed interface landslides occur frequently in Northwest China, and the critical conditions for their initiation are closely related to the dynamic evolution of the shear strength at the contact interface. To investigate the initiation mechanisms of such landslides, this study takes the slip zone soils of typical loess-red bed interface landslides as research objects. Through ring shear tests and scanning electron microscopy (SEM) analyses conducted on three types of samples (pure loess, pure red bed, and loess-red bed composites) the study systematically evaluates the evolution of shear strength and microscopic structures under varying normal stresses, water contents, and shear rates. The results show that: (1) The peak and residual shear strengths of all three soil types increase markedly with increasing normal stress. The peak shear strength exhibits a unimodal variation with water content and shows a slight increasing trend with increasing shear rate. (2) The loess-red bed composite samples are most sensitive to water content changes, with the most significant strength attenuation. Their brittleness index reaches 15.6%, significantly higher than that of the single-component soils, indicating a stronger tendency toward strain-softening failure. (3) Microscopic observations reveal that increasing water content weakens cementation of clay particles, and the enhanced lubricating of flaky minerals promotes the formation of smooth shear surfaces, leading to a significant reduction in interfacial shear strength. Integrating field investigations with laboratory tests, the initiation mechanism of loess-red bed interface landslides is clarified. The hydraulic weakening effect of the loess-red bed interface is identified as the core mechanism controlling slope instability.

In recent years, the continuous advancement of engineering construction in China's Loess Plateau region has resulted in numerous high-fill and deep-excavation loess-red layer composite slopes (HDLS). Rainfall infiltration, a primary factor influencing slope stability, has markedly increased the likelihood of landslide disasters. To investigate the primary sliding failure mechanism of HDLS, a series of physical model tests and numerical simulation analyses were conducted on a loess-red layer composite slope in Zhongliang Town, Tianshui City. The study systematically examined the infiltration characteristics, deformation response, crack extension, and failure patterns of HDLS under rainfall conditions. The results demonstrate that (1) analysis of monitoring data from multi-sensors and real-time slope imagery revealed that under constant rainfall intensity, the slope failure process accelerates significantly with increasing slope height and angle. Furthermore, the slope shoulder exhibits greater susceptibility to large-scale sliding deformation due to enhanced rainwater infiltration. (2) Numerical simulations using Geo-studio reveal that the slope safety factor decreases significantly under fill and excavation conditions. The safety factor of high-fill slopes exhibits a slower decline with prolonged rainfall duration. (3) Rainfall erosion readily develops gully sliding ways on fill slopes, with tension cracks in the mid-slope region serving as primary infiltration channels that constitute the dominant failure reason. As the excavated slope angle increases, both the slope toe and mid-slope experience significantly enhanced unit water flow, intensifying surface runoff erosion, and the failure pattern exhibits progressive upward development originating from the slope toe to the upper slope. This paper investigates the hydrologic response characteristics and sliding deformation mechanisms of HDLS under various working conditions. The findings provide a significant reference value for protecting such high and steep slopes.

Mudflows are a prevalent hazard in loess regions, yet their initiation mechanisms and critical conditions remain incompletely understood. Through flume experiments, we investigated how rainfall intensity, slope angle, and dry density control the initiation of loess mudflows. The results show that the initiation modes of loess mudflows are mainly static liquefaction initiation, landslide transformation initiation, and erosion-expansion initiation, with landslide transformation initiation being the most common. The initiation time increases with dry density but decreases then increases with slope and rainfall intensity, revealing a slope angle of 35°and rainfall intensity of 90 mm/h to be more prone to trigger large mudflow. Quantitative analysis of crack evolution reveals a direct correlation between crack density and failure mechanisms. The critical water content for this transition (23.99–36.02

Accumulation landslide induced by rainfall is one of the most important types of geological disasters in the Qinling-Bashan Mountains, China. In recent years, nearly one hundred accumulation landslides have caused significant casualties and serious economic losses. To better understand the response mechanism of sliding zone soil to such landslides under rainfall infiltration conditions, the landslide in Zhashui County, Shaanxi Province, was taken as a typical case. The field investigation, ring shear test and creep test were carried out using sliding zone soil. Combined with the laboratory tests results, the landslide was numerically simulated and analysed. Laboratory test results show that the increase in moisture content leads to a reduction in the shear strength of the sliding zone soil, promoting slope creep and accelerating the deformation of the slope. Numerical simulation results for two typical rainfall infiltration scenarios, short-duration heavy rainfall and long-duration weak rainfall, indicate that the failure type of accumulation landslide is a creep-slide failure, and the damage degree of the heavy rainfall to the slope is greater than that of the weak rainfall. According to the results of the field investigation and numerical simulation, we find that the mechanical behaviour of the sliding zone soil controls the failure mode of the accumulation landslide in Qinling-Bashan Mountains. This kind of landslide has roughly experienced three failure stages: the early disaster-breeding stage, the interim accelerated deformation stage and the anaphase instability failure stage.

The occurrence of landslides is often a complex dynamic evolution process, characterized by multiple slide-stop-slide cycles. Each state transition may be associated with changes in the mechanical state and structure of the sliding zone soil, significantly influencing the overall stability of the landslide. Taking the sliding zone soil of Ertaizi gully landslide as test material, a series of repeated ring shear tests were conducted under different moisture contents (Mc), normal stresses (σn), and the number of shear cycles (NSC) to investigate the mechanical strength characteristics of the sliding zone soil during the process of repetitive motion landslides movement. The results indicate: (1) Mc critically influences the soil shear strength in the deep sliding zone, while increasing NSC diminishes the strain-softening phenomenon of soil. Additionally, the attenuation effect of NSC against shear strength actually weakens the cohesion of the soil. (2) An increase in Mc reduces soil looseness near the shear plane, while an increase in σn expands relatively flat region adjacent to the shear plane. Under shearing, soil particles align along the shearing direction, exhibiting a preferred orientation within 0 30°. (3) Although there is a healing mechanism at the shear plane, repetitive motion landslides remain highly hazardous. This research offers a theoretical basis for comprehending the effects of Mc, σn, and NSC on the mechanical strength and microstructural deformation properties of soil. Additionally, it provides guidance for predicting the stability of repetitive motion landslides.