Gas station sites pose potential risks of soil and groundwater contamination, which not only threatens public health and property but may also damage the assets and reputation of businesses and government entities. Given the complex nature of soil and groundwater contamination at gas station sites, this study utilizes field data from basic and environmental information, maintenance information for tank and pipeline monitoring, and environmental monitoring to develop machine learning models for predicting potential contamination risks and evaluating high-impact risk factors. The research employs three machine learning models: XGBoost, LightGBM, and Random Forest (RF). To compare the performance of these models in predicting soil and groundwater contamination, multiple performance metrics were utilized, including Receiver Operating Characteristic (ROC) curves, Precision-Recall graphs, and Confusion Matrix (CM). The Confusion Matrix analysis revealed the following results: accuracy of 85.1-87.4 %, precision of 86.6-88.3 %, recall of 83.0-87.2 %, and F1 score of 84.8-87.8 %. Performance ranking across all metrics consistently showed: XGBoost > LightGBM > RF. The area under the ROC curve and precision-recall curve for the three models were 0.95 (XGBoost), 0.94 (LightGBM), and 0.93 (RF), respectively. While all three machine learning approaches demonstrated satisfactory predictive capabilities, the XGBoost model exhibited optimal performance across all evaluation metrics. This research demonstrates that properly trained machine learning models can serve as effective tools for environmental risk assessment and management. These findings have significant implications for decision-makers in environmental protection, enabling more accurate prediction and control of contamination risks, thereby enhancing the preservation of ecological systems, public health, and property security.

Debris flows are destructive mass movements that pose multifaceted challenges with profound social and environmental implications in the Western Himalayas. For precise modeling and flow behavior prediction, it is essential to understand the rheological characteristics of debris flow material. In the current study, rheological characteristics like yield stress and viscosity were determined by a series of lab tests using a parallel plate setup in a rheometer. An optimized sampling approach created the reconstituted soil samples of finer particles to change the solid volume concentration and volumetric water content (w/c). Later, the feature importance of finer particles in debris flow rheology was determined using a machine learning regressor. Non-Newtonian behavior was shown by each composition and was similar to Herschel-Bulkley's rheological model. The eXtreme Gradient Boosting (XGBoost) regression model was developed for rheological parameters with robust model fitting with R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}{2}$$\end{document} = 0.90 for yield stress and R2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}{2}$$\end{document} = 0.94 for viscosity. The model helped in understanding the sensitivity of rheological parameters with solid constitutents of debris flows. The findings showed that water content and silt concentration substantially impacted the debris flow's rheology. The yield stress was more dominated by silt followed by fine sand, whereas water content influenced the viscosity more than any solid concentration. The flow behavior was also affected by the distribution of grain sizes, with finer particles exhibiting higher viscosity and shear stress than coarser particles. These results enhance understanding of debris flow rheology and highlight the complex interplay between geohazards and sustainable development.

Pipes and tunnels are prone to longitudinal deformation under normal faulting. Predicting the active length, defined as the major deformed length, is crucial for the seismic design of pipes and tunnels. However, an accurate prediction method is hard to develop due to a limited number of experimental data. In this study, a large number of numerical simulations are conducted based on a three-dimensional beam-spring model validated by centrifuge tests. The results are fed to XGBoost models to develop a robust prediction method. For ease of application, only the friction angle of soil, burial depth, diameter, and thickness of pipes or tunnels are incorporated as features. A hyperparameter tuning approach integrating grid search and Bayesian optimization was employed in the training process to establish optimal models with comparatively low complexity and high accuracy. A comparison of predictions from the XGBoost models and curves fitted on relative structure-soil stiffness demonstrates that XGBoost models are much superior. The effects of each feature on the predictions were analyzed by employing the SHAP method. The proposed XGBoost models can effectively and efficiently predict the active length of pipes and tunnels with minimal inputs.

The eastern of West Bengal grapples with limited surface water availability in its hard rock terrain, compounded by a semi-arid climate, variable rainfall, and a plateau topography, prompting communities to adapt groundwater water-use practices, leading to unsustainable extraction and misuse. Thus, the novel objective of the present research was to produce groundwater potential maps by comparing machine learning techniques with a Fuzzy MCDM model using specific field-based conditioning factors. In the first step, 285 wells were identified, of which 70 percent were used for training and 30 percent for the validation of the models. Secondly, field-based conditioning factors including, longitudinal conductance (SC), longitudinal resistance (rho l), transverse resistance (TR), coefficient of electrical anisotropy (lambda), resistivity of formation (rho m), fracture porosity (phi f), reflection coefficients (r), hydraulic conductivity (K), transmissivity(Tr), bulk density, porosity, permeability, soil moisture content and water holding capacity were used to analyze the association between these conditioning factors and groundwater occurrences. In the following steps, the XGBoost, Random Forest, and Na & iuml;ve Bayes models were executed using the training dataset, and factor weights were calculated using Fuzzy Analytical Hierarchy Process of Extent analysis method. To validate and compare the performance of four models, ROC curves, AUCs, MCAs, and correlation plots were used. In general, all four models were successful in evaluating the potential of groundwater occurrences. The predictive capability of the XGBoost techniques with the highest AUC values (0.79) and the highest correlation value (0.78) is superior to those of other machine learning and MCDM models. Geophysical survey revealed that transmissivity and hydraulic conductivity of the aquifer of the river basin range from 1.55 to 440.11 m/day and 10.15-2253 m(2)/day, indicating a moderate to good hydrodynamic potential. Planners and engineers can use such groundwater potential maps to manage water resources effectively.

Braced excavation is a common practice in underground construction for buildings and tunneling in urban areas. To determine the stability and failure behavior of the surrounding soil around the excavation, ensuring that it does not collapse and maintains safety, this study focuses on investigating the undrained stability of a fully supported 3D rectangular excavation with an embedded wall in anisotropic clay. The analysis is conducted through three-dimensional finite element limit analysis (3D FELA). To model the failure behavior of soils, the anisotropic undrained shear strength (AUS) failure criterion is employed in FELA. The factors influencing excavation failure are categorized into four dimensionless variables: the anisotropic strength ratio (re), the embedded rigid wall ratio (D/H), the aspect ratio (A/B), and the depth ratio (H/B). The solutions of the stability number (N), as well as the failure mechanism of 3D excavations, are derived from both the lower bound (LB) and upper bound (UB) approaches. This paper also introduces a novel machine learning approach for predicting the stability of supported rectangular excavations in anisotropic clays. Moreover, this study investigates the potential of using an advanced machine learning model which is XGBoost technique. The gathered evidence indicates that XGBoost demonstrates remarkable precision in forecasting the stability of supported excavation in anisotropic clays.

Concrete structures in saline soil regions are prone to degradation due to chloride and sulfate erosion, compounded by the concurrent infiuences of drying, high and low temperatures, and freeze-thaw cycles. This study establishes a simulation test system for complex saline soil environments, integrating findings from real-world environmental investigations. The investigation focused on the degradation mechanism of concrete under the combined impacts of dry-wet and high-low temperature cycles, coupled with composite salt erosion. Additionally, the impacts of water-cement ratio, fiy ash content, and basalt fiber content on concrete's mechanical properties and ion erosion resistance were analyzed. The alterations in the internal pore structure of corroded concrete were examined through nuclear magnetic resonance (NMR) technology. Utilizing the XGBoost algorithm, a predictive model for chloride and sulfate ion concentrations in concrete, under the combined infiuence of dry-wet and high-low temperature cycles, coupled with composite salt erosion, was developed. The findings reveal that the rate of concrete deterioration is gradually accelerating under the combined erosion to dry-wet cycles, high-low temperature cycles, and composite salt. Optimal fiy ash and basalt fiber dosages for corrosion resistance are determined to be 10% and 0.10%, respectively. During advanced erosion stages, concrete porosity, capillary and macropore volume fractions increase, while gel pore volume fraction declines significantly. The XGBoost-based chloride and sulfate concentration prediction model demonstrates strong agreement with experimental measurements, yielding correlation indices of R2 = 0.98 and 0.97, respectively. Interpretation results obtained using SHAP from the machine learning model align with experimental outcomes.

Thermal conductivity is a key soil property widely used for agricultural production, land surface processing research, and geothermal resource development, among others. Although the rapid and accurate determination of soil thermal conductivity (A) has been a hot topic in recent years, there is still no unified model for the different soil types of soil. Furthermore, the lack of data on thermal conductivity and soil properties leads to errors in parametric models of thermal conductivity. In order to overcome the data shortage, a comprehensive lambda dataset of 2972 items was established and 10 influential parameters on thermal conductivity were identified in this study. Based on this, an empirical comparison was made between four classical parametric models and nine machinelearning models with and without an intelligent optimization algorithm was carried out. Of all the methods, the ensemble machine-learning methods perform better in lambda simulations. The XGBoost model has better simulation accuracy and generalization capability. Soil moisture properties are the key parameters in performing lambda simulations, while the soil texture-related properties such as bulk density and solid thermal conductivity, along with the sand content, also play an important role. The results of this study can provide basic thermal conductivity data and a parameterization scheme for referencing in land surface processing research.