Tree destruction induced by heavy rainfall, an overlooked type of forest degradation, has been exacerbated along with global climate change. On the Chinese Loess Plateau, especially in afforested gully catchments dominated by Robinia pseudoacacia, destructive rainfall events have increasingly led to widespread forest damage. Previous study has manifested the severity of heavy rainfall-induced tree destruction and its association with topographic change, yet the contributions of tree structure and forest structure remain poorly understood. In this study, we quantified the destroyed trees induced by heavy rainfall using light detection and ranging (LiDAR) techniques. We assessed the influence of tree structure (tree height, crown diameter, and crown area), forest structure (tree density, gap fraction, leaf area index, and canopy cover), and terrain parameters (elevation, slope, and terrain relief) using machine learning models (random forest and logistic regression). Based on these, we aimed to clarify the respective and combined contributions of structural and topographic factors to rainfall-induced tree destruction. Key findings revealed that when considered in isolation, greater tree height, crown diameter, crown area, leaf area index (LAI), and canopy cover suppressed tree destruction, whereas higher gap fractions increased the probability of tree destruction. However, the synergistic increases of tree structural factors (tree height, crown diameter, and crown area) and forest structural factors (LAI and canopy cover) significantly promoted tree destruction, which can counteract the inhibitory effect of terrain on destruction. In addition, increases in tree structure or canopy density (LAI and canopy cover) also increased the probability of tree destruction at the same elevation. Our findings challenge conventional assumptions in forest management by demonstrating the interaction of tree structure and canopy density can significantly promote tree destruction during heavy rainfall. This highlights the need to avoid overly dense afforestation in vulnerable landscapes and supports more adaptive, climate-resilient restoration strategies.

Floods are devastating natural disasters causing significant damage worldwide, especially in southern Latin America, where recurrent river floods lead to severe impacts. This study proposes an innovative flood modelling approach using a naive Bayes classifier to simulate flood extents at a regional scale, incorporating spatial and temporal variability. Using 12 features, including topography, soil properties, precipitation and discharge, the model was trained with multiple flood events, avoiding sampling limitations and evaluating optimal pre-processing strategies for continuous data. The predictive capacity resulted in high performance metrics, with temporal validation accuracy (AC) up to 0.98 and a critical success index (CSI) of 0.58, and spatial validation achieved an AC up to 0.97 and CSI of 0.56, outperforming the hydrodynamic model by 65%. A reduced model with significant features improved computational efficiency and achieved a CSI exceeding 0.60. This practical tool supports flood risk management and enhances resilience in vulnerable regions.

Moisture accumulation within road pavements, particularly in unbound granular materials with or without thin sprayed seals, presents significant challenges in high-rainfall regions such as Queensland. This infiltration often leads to various forms of pavement distress, eventually causing irreversible damage to the pavement structure. The moisture content within pavements exhibits considerable dynamism and directly influenced by environmental factors such as precipitation, air temperature, and relative humidity. This variability underscores the importance of monitoring moisture changes using real-time climatic data to assess pavement conditions for operational management or incorporating these effects during pavement design based on historical climate data. Consequently, there is an increasing demand for advanced, technology-driven methodologies to predict moisture variations based on climatic inputs. Addressing this gap, the present study employs five traditional machine learning (ML) algorithms, K-nearest neighbors (KNN), regression trees, random forest, support vector machines (SVMs), and gaussian process regression (GPR), to forecast moisture levels within pavement layers over time, with varying algorithm complexities. Using data collected from an instrumented road in Brisbane, Australia, which includes pavement moisture and climatic factors, the study develops predictive models to forecast moisture content at future time steps. The approach incorporates current moisture content, rather than averaged values, along with seasonality (both daily and annual), and key climatic factors to predict next step moisture. Model performance is evaluated using R2, MSE, RMSE, and MAPE metrics. Results show that ML algorithms can reliably predict long-term moisture variations in pavements, provided optimal hyperparameters are selected for each algorithm. The best-performing algorithms include KNN (the number of neighbours equals to 15), medium regression tree, medium random forest, coarse SVM, and simple GPR, with medium random forest outperforming the others. The study also identifies the optimal hyperparameter combinations for each algorithm, offering significant advancements in moisture prediction tools for pavement technology.

This study investigates the stabilization of lateritic soil through partial replacement of cement with flue gas desulfurization (FGD) gypsum, aiming to enhance its engineering properties for pavement subgrade applications. Lateritic soils are known for their high plasticity and low strength, which limit their utility in infrastructure. To address these challenges, soil specimens were treated with varying cement contents (3%, 6%, 9%) and FGD gypsum additions (1%-6%). Laboratory tests were conducted to evaluate plasticity, compaction, permeability, unconfined compressive strength (UCS), California Bearing Ratio (CBR), and fatigue behaviour. The optimal mix 6% cement with 3% FGD gypsum demonstrated significant improvements: UCS increased by over 110% after 28 days, permeability reduced by 26%, and soaked CBR improved by 56% compared to untreated soil. Additionally, fatigue life showed remarkable enhancement under cyclic loading, indicating increased durability for high-traffic applications. To support predictive insights, machine learning models including Decision Tree, Random Forest, and Multi-Layer Perceptron (MLP) were trained on 168 data samples. The MLP and Random Forest models achieved high prediction accuracy (R2 approximate to 0.98), effectively capturing the non-linear interactions between mix proportions and UCS. SHAP (SHapley Additive exPlanations) analysis identified curing duration as the most influential factor affecting strength development. This integrated experimental-computational approach not only validates the feasibility of using FGD gypsum in sustainable soil stabilization but also demonstrates the effectiveness of machine learning in predicting key geotechnical parameters, reducing reliance on extensive laboratory testing and promoting data-driven pavement design.

Roads in places with seasonal frost undergo several freeze-thaw (F-T) cycles annually, resulting in variable degrees of deterioration in the mechanical properties of the subgrade. To methodically investigate the mechanical properties of subgrade clay during freeze-thaw cycles and to develop a precise constitutive model, triaxial tests were conducted under the most unfavorable soil conditions. The studies indicate that the degrading impact of the freeze-thaw cycle on the mechanical characteristics of the soil predominantly transpires during the initial freeze-thaw cycle. Soil strength reaches its minimum after the third freeze-thaw cycle, followed by a slight increase, and ultimately stabilizes between the fifth and seventh cycles. The maximum strength reduction at confining pressures of 100 kPa, 200 kPa, and 300 kPa was 39%, 37%, and 33%, respectively. As confining pressure escalates, the reduction in soil strength lessens. The soil demonstrates differing degrees of degradation following F-T cycles at both high and low compaction levels, with the degradation becoming increasingly evident as compaction intensifies. Utilizing the experimental database, a genetic algorithm (GA) enhanced backpropagation neural network (BPNN) model (GA-BPNN) and a BP-aided Duncan-Chang (D-C) model were developed to forecast the mechanical properties of freeze-thaw clay. The R2 values for the two models on the test set were 0.995 and 0.967, respectively. The efficacy of these two models demonstrates that machine learning can attain commendable outcomes in extensive data structures (total stress-strain curve) as well as exhibit superior performance in limited data (model parameters) while developing the constitutive model of soil.

Soil organic carbon (SOC) in the active layer (0-2 m) of the Tibetan Plateau (TP) permafrost region is sensitive to climate change, with significant implications for the global carbon cycle. Environmental factors-including parent material, climate, vegetation, topography, soil, and human activities-inevitably drive SOC variations. However, vegetation and climate are likely the two most influential factors impacting SOC variations. To test this hypothesis, we conducted experiments using 31 environmental variables combined with the recursive feature elimination (RFE) algorithm. These experiments showed that RFE retained all vegetation variables [Land cover types (LCT), normalized difference vegetation index (NDVI), leaf area index (LAI), and gross primary productivity (GPP)] as well as two climate variables [Moisture index (MI) and drought index (DI)], supporting our hypothesis. We then analyzed the relationship between SOC and the retained vegetation and climate variables using random forest (RF), Shapley additive explanations (SHAP), and GeoDetector models to quantify the independent and interactive drivers of SOC distribution and to identify the optimal conditions for SOC accumulation. The RF model explained 68% and 42% of the spatial variability in SOC at depths of 0-1 m and 1-2 m, respectively, with SOC stocks higher in the southeast and lower in the northwest. Additionally, SOC stock at 0-1 m was significantly higher (p 0.05). Spearman correlation coefficients results indicated that NDVI, LAI, GPP, and MI had highly significant positive correlations with SOC (p < 0.01), whereas DI had a highly significant negative correlation with SOC (p < 0.01). SHAP analysis revealed environmental thresholds for SOC variations, with notable shifts at NDVI (0.40), LAI (7), GPP (250 g C m(-)(2) year(-)(1)), MI (0.40), and DI (0.50). The spatial distribution of these thresholds aligns with the 400 mm equivalent precipitation line. Additionally, GeoDetector results emphasized that interactions between climate and vegetation factors enhance the explanatory power of individual variables on SOC variations. The swamp meadow type, with an NDVI range of 0.73-0.84, LAI range of 11.06-15.94, and MI range of 0.46-0.56, was identified as the most favorable environment for SOC accumulation. These findings are essential for balancing vegetation and climate conditions to sustain SOC levels and mitigate climate change-driven carbon release.

The accelerated warming in the Arctic poses serious risks to freshwater ecosystems by altering streamflow and river thermal regimes. However, limited research on Arctic River water temperatures exists due to data scarcity and the absence of robust methodologies, which often focus on large, major river basins. To address this, we leveraged the newly released, extensive AKTEMP data set and advanced machine learning techniques to develop a Long Short-Term Memory (LSTM) model. By incorporating ERA5-Land reanalysis data and integrating physical understanding into data-driven processes, our model advanced river water temperature predictions in ungauged, snow- and permafrost-affected basins in Alaska. Our model outperformed existing approaches in high-latitude regions, achieving a median Nash-Sutcliffe Efficiency of 0.95 and root mean squared error of 1.0 degrees C. The LSTM model learned air temperature, soil temperature, solar radiation, and thermal radiation-factors associated with energy balance-were the most important drivers of river temperature dynamics. Soil moisture and snow water equivalent were highlighted as critical factors representing key processes such as thawing, melting, and groundwater contributions. Glaciers and permafrost were also identified as important covariates, particularly in seasonal river water temperature predictions. Our LSTM model successfully captured the complex relationships between hydrometeorological factors and river water temperatures across varying timescales and hydrological conditions. This scalable and transferable approach can be potentially applied across the Arctic, offering valuable insights for future conservation and management efforts.

Building structures on clayey soil presents unique challenges to geotechnical engineers due to the inherent variability in clayey soil consistency. Understanding engineering properties of clayey soils is essential for accurate geotechnical design and the prevention of potential issues such as settlement and instability. The current study provides crucial insights for geotechnical assessments and engineering solutions in the area, highlighting key soil properties that affect the classification of clayey consistency. Advanced machine learning (ML) models were employed to predict the in situ clay consistency, a vital parameter for evaluating the deformation resistance of clayey soils under structures. The ML predictions are based on nine features representing the physical and mechanical properties of the clay, which are easily determined through laboratory and field evaluations. A dataset comprising 173 samples is compiled, which extracted from Nile Delta in Egypt, incorporating data on the basic properties of the soils to train and test several ML classification algorithms. The classification models, including logistic regression, k-nearest neighbors, support vector machine, random forest, and gradient boosting classifiers, are evaluated using metrics such as accuracy, sensitivity, specificity, and F1-score. The results demonstrate that the gradient boosting classifier model exhibits the highest accuracy in predicting clay class, achieving 97% and 86% accuracy for the training and testing datasets, respectively. These findings offer a valuable framework for efficiently and cost-effectively classifying clays, assisting geotechnical engineers in making informed decisions about foundation design and construction on clayey soils. Additionally, the study establishes equations to predict the undrained shear strength of clayey soil based on its basic properties, providing a practical and accurate method for estimating soil strength characteristics. These contributions enhance the understanding and management of clayey soil behavior in geotechnical engineering, offering essential guidance for foundation design and construction projects in clayey soil regions.

In complex physical systems, conventional differential equations fall short in capturing non-local and memory effects. Fractional differential equations (FDEs) effectively model long-range interactions with fewer parameters. However, deriving FDEs from physical principles remains a significant challenge. This study introduces a stepwise data-driven framework to discover explicit expressions of FDEs directly from data. The proposed framework combines deep neural networks for data reconstruction and automatic differentiation with Gauss-Jacobi quadrature for fractional derivative approximation, effectively handling singularities while achieving fast, high-precision computations across large temporal/spatial scales. To optimize both linear coefficients and the nonlinear fractional orders, we employ an alternating optimization approach that combines sparse regression with global optimization techniques. We validate the framework on various datasets, including synthetic anomalous diffusion data, experimental data on the creep behavior of frozen soils, and single-particle trajectories modeled by L & eacute;vy motion. Results demonstrate the framework's robustness in identifying FDE structures across diverse noise levels and its ability to capture integer-order dynamics, offering a flexible approach for modeling memory effects in complex systems.

The laying of the underground pipeline in the same ditch has caused great challenges to the attractive transportation mode of hydrogen mixed with natural gas pipeline in service. The tendency to damage of hydrogen to steel increases the possibility of flammable and explosive gas entering underground engineering significantly. A leakage monitoring method for buried hydrogen-doped natural gas pipeline based on vibration signals with machine learning is proposed. Firstly, the distributed vibration sensor captures the multisource vibration signals propagating in the soil. An optimal combination of wavelet basis functions, decomposition level, and threshold parameters is selected carefully for signal denoising and accurate extraction of leakage-generated signals. Then the characteristics extracted in different frequency bands are investigated with other influencing factors, including the hydrogen-doping ratio, which affects the propagation speed of the pressure wave. The unique characteristics of vibration signal generated by pipeline leakage are extracted. On this basis, combined with the high efficiency of machine learning recognition model, a leakage monitoring model for buried hydrogen-doped natural gas pipeline is established, which achieves a 2.01 % false alarm rate at a maximum positioning distance of 70 cm. It has been successfully applied to the leak detection and location of buried hydrogen-doped natural gas pipelines, which can significantly improve the safety and reliability of underground pipeline system engineering.