Landslide susceptibility assessment is critical for hazard mitigation and land-use planning. This study evaluates the impact of two different non-landslide sampling methods—random sampling and sampling constrained by the Global Landslide Hazard Map (GLHM)—on the performance of various machine learning and deep learning models, including Naïve Bayes (NB), Support Vector Machine (SVM), SVM-Random Forest hybrid (SVM-RF), and XGBoost. The study area is a 2 km buffer zone along the Duku Highway in Xinjiang, China, with 102 landslide and 102 non-landslide points extracted by aforementioned sampling methods. Models were tested using ROC curves and non-parametric significance tests based on 20 repetitions of 5-fold spatial cross-validation data. GLHM sampling consistently improved AUROC and accuracy across all models (e.g., AUROC gains: NB +8.44, SVM +7.11, SVM–RF +3.45, XGBoost +3.04; accuracy gains: NB +11.30%, SVM +8.33%, SVM–RF +7.40%, XGBoost +8.31%). XGBoost delivered the best performance under both sampling strategies, reaching 94.61% AUROC and 84.30% accuracy with GLHM sampling. SHAP analysis showed that GLHM sampling stabilized feature importance rankings, highlighting STI, TWI, and NDVI as the main controlling factors for landslides in the study area. These results highlight the importance of hazard-informed sampling to enhance landslide susceptibility modeling accuracy and interpretability.
Snowdrift, as a natural disaster, constantly compromises railway traffic by affecting how snow accumulates on the subgrade. This paper establishes a unified set of similarity criteria for wind tunnel testing, using viscous silica sand to simulate snow particles. By employing a geometric scale model (1:30) and similarity criteria (size, motion, dynamics, accumulation patterns, and time scales), it systematically investigates the evolution patterns of wind-induced snow accumulation on two types of roadbed structures: embankments and excavations. This study also evaluates the effectiveness of snow fences, proposing optimized placement distances and quantifying the effects of snow accumulation platform width. The results showed the following: (1) Snow on embankments has a “U”-shaped distribution, with the lowest wind speed (<0.5 m/s) and maximum accumulation at the leeward slope’s foot. In excavations, snow forms an “M”-shaped distribution, with significantly reduced wind speeds (<1 m/s) on the accumulation platform. (2) Snow fences effectively manage snow placement by lowering wind speed (below 1 m/s). A single-row snow fence with a porosity of 50% and a height of 3 m performs best when placed at seven times its height (7 H) from the slope’s toe. (3) A 5 m snow accumulation platform in excavations reduces surface snow accumulation (distribution coefficient drops to 1.6), outperforming scenarios without a platform (coefficient of 2.0). These findings contribute to the prevention and control of snowdrift disasters along railway lines in cold regions. They offer practical guidance for optimizing snow fence configurations, while also laying a foundation for future improvements in experimental accuracy through advanced techniques such as PIV and real-snow testing.
Landslide susceptibility assessment is critical for hazard mitigation and land-use planning. This study evaluates the impact of two different non-landslide sampling methods—random sampling and sampling constrained by the Global Landslide Hazard Map (GLHM)—on the performance of various machine learning and deep learning models, including Naïve Bayes (NB), Support Vector Machine (SVM), SVM-Random Forest hybrid (SVM-RF), and XGBoost. The study area is a 2 km buffer zone along the Duku Highway in Xinjiang, China, with 101 landslide and 101 non-landslide points extracted by aforementioned sampling methods. Models were tested using ROC curves and non-parametric significance tests based on 20 repetitions of 5-fold spatial cross-validation data. GLHM sampling consistently improved AUROC and accuracy across all models (e.g., AUROC gains: NB +8.44, SVM +7.11, SVM-RF +3.45, XGBoost +3.04; accuracy gains: NB +11.30%, SVM +8.33%, SVM-RF +7.40%, XGBoost +8.31%). XGBoost delivered the best performance under both sampling strategies, reaching 94.61% AUROC and 84.30% accuracy with GLHM sampling. SHAP analysis showed that GLHM sampling stabilized feature importance rankings, highlighting STI, TWI and NDVI are the main controlling factors for landslides in the study area. These results highlight the importance of hazard-informed sampling to enhance landslide susceptibility modeling accuracy and interpretability.
Aeolian sand is widely distributed in arid and semi-arid regions, but its loose structure and low cohesion limit its direct engineering use. This study investigates the stabilization of aeolian sand using alkali-activated steel slag—fly ash system. Single-factor tests and an L25 (5 4 ) orthogonal design were conducted to evaluate the effects of steel slag content, fly ash content, NaOH dosage, and moisture content on shear strength and unconfined compressive strength. SEM, XRD, and FTIR analyses were further used to clarify the solidification mechanism. The results show that steel slag content is the dominant factor controlling shear strength and strength parameters, followed by NaOH dosage and fly ash content, while moisture content has the weakest influence. The optimal mixture was 12% steel slag, 10% fly ash, 4% NaOH, and 12% moisture. Under this condition, cohesion and internal friction angle reached 198.2 kPa and 51.1°, respectively, and UCS reached 2450 kPa at 7 d and 4860 kPa at 28 d. The strength improvement was mainly attributed to the formation of alkali-activated gels, pore filling, and enhanced interparticle bonding.
Snow drift disasters pose a serious threat to the safe operation of railway engineering. Current research on the characteristics of railway snow drift disasters and the role of snow fences in their prevention often relies on indoor wind tunnel experiments, which overlook the influence of factors such as temperature, humidity, and climate on snow particle transport. To realistically reproduce the forces acting on snow particles and their movement patterns, this study, based on field monitoring results, employed a mobile wind tunnel simulation device and applied similarity criteria to conduct scaled outdoor wind tunnel tests on-site. The study examined the impact of the distance between snow fences and the railway subgrade on snow accumulation distribution. It also analyzed how different fence-subgrade distances affect wind fields and surface shear velocity around the subgrade. Furthermore, numerical simulations were used to investigate the influence of snow fence porosity, the distance between snow fences and the cutting slope, and fence height on wind and snow field distributions near the subgrade. Optimal design parameters for snow fences were proposed accordingly. The results show that outdoor in-situ wind tunnel experiments can realistically replicate field engineering environments (including appropriate temperature and humidity), allowing for more scientific testing of the geometry, snow particle saltation processes, particle trajectories, time scales, and deposition patterns during snow drift disasters affecting the subgrade. For railway sections prone to snow drift disasters, snow fences with 50% porosity, a height of 3 meters, and placed 50 meters from the cutting slope provide optimal protection. These findings offer theoretical support for the prevention and control of snow drift disasters and for the safety-oriented design of railway subgrade engineering in high-risk areas.