Search Results (2,787)

Oil palm plantations play an important role in agricultural production and carbon storage in tropical regions. The accurate estimation of aboveground carbon (AGC) is essential for sustainable plantation management, climate change mitigation, and carbon monitoring. Although field measurements provide reliable estimates, they are often time-consuming, labor-intensive, and costly, particularly over large plantation areas. Recent advances in remote sensing and machine learning offer efficient alternatives for AGC estimation using satellite imagery. In this study, we developed a machine learning framework for AGC estimation in oil palm plantations using Sentinel-2 multispectral imagery and Sentinel-1 synthetic aperture radar (SAR) data. Field measurements were integrated with spectral variables, vegetation indices, and SAR-derived parameters extracted from satellite data. A hybrid feature selection approach combining Pearson correlation, mutual information and mRMR was used to identify the most relevant variables. Six machine learning algorithms were evaluated, including Linear Regression, Random Forest, XGBoost, Gradient Boosting, LightGBM, and Extra Trees. Because the 160 observations comprise sixteen 10 m × 10 m grid cells nested within ten 40 m × 40 m field plots, model performance was assessed with leave-one-plot-out cross-validation: all sixteen cells of a plot were held out together, and the hybrid feature selection was repeated inside every fold using only that fold’s training plots. Performance was measured on pooled out-of-fold predictions using R², root mean squared error (RMSE), and average absolute relative error (AARE%). Under this spatially independent design the combined Sentinel-1 + Sentinel-2 dataset gave the highest accuracy (R² = 0.7950, RMSE = 4.14 t C ha⁻¹, AARE = 33.53%), followed by Sentinel-1 alone (R² = 0.7631, RMSE = 4.46 t C ha⁻¹) and Sentinel-2 alone (R² = 0.6108, RMSE = 5.71 t C ha⁻¹). Linear Regression and Extra Trees were the most robust models, whereas the boosted ensembles did not generalize to unseen plots. Repeating the evaluation with an ungrouped random split of the same data inflated R² by up to 0.70, showing that a large part of the accuracy obtainable under that design reflects within-plot spatial autocorrelation rather than predictive skill. These findings indicate that optical-SAR imagery combined with machine learning can provide useful AGC estimates in oil palm plantations, and that spatially independent validation is essential for reporting them honestly. The proposed framework can be used to support plantation-scale carbon mapping, monitoring, and carbon stock assessment, subject to further calibration and independent validation across additional plantations. Full article

Mobile health (mHealth) apps increasingly capture the physiological, biomechanical, and psychological variables involved in sport training, but the evidence remains fragmented across single-domain reviews, leaving practitioners without a consolidated basis for selecting and deploying these tools across the training process. This scoping review aimed to identify and characterize research on mHealth apps in sport training, focusing on their performance testing, training load and recovery monitoring, technical and skill development, injury screening and prevention, and athlete self-management. It also synthesized evidence regarding their applications, intended purposes, technical characteristics, and the evidence supporting their effectiveness. Five databases (PubMed, Scopus, Web of Science, SPORTDiscus, Embase) were searched from inception to July 2026 for journal articles reporting original empirical data on app research on the sport training process in athletes. Studies involving only the promotion of physical activity or lacking human-subject testing, including commercially available apps without supporting research on their effectiveness, were excluded. Findings were synthesized narratively, and methodological quality was appraised with the Mixed Methods Appraisal Tool. Of 9476 records identified, 111 studies met the inclusion criteria and were inductively classified into ten application categories: sport skill training (n = 26), performance measurement (n = 20), vertical jump measurement (n = 18), self-reported monitoring (n = 12), physiological measurement (n = 12), musculoskeletal screening (n = 10), psychological intervention (n = 5), nutrition (n = 3), anthropometric and maturation screening (n = 3), and tactical and match analysis (n = 2). Most apps relied on built-in smartphone sensors or no sensing at all and used manual or deterministic computation; processing location went unreported in 74.8% of studies, which reflects a reporting gap rather than an architectural profile of the field, and reported that AI or machine learning labels did not track with actual method disclosure. Validation and reliability designs dominated the evidence base (52%), while randomized or controlled effectiveness trials were rare (10%). Apps generally showed good relative validity but limited absolute accuracy against criterion instruments, and wherever apps were deployed longitudinally, adherence rather than accuracy determined their real-world value. mHealth apps now support nearly every stage of sport training and can substitute for laboratory instruments in select, validated use cases, including video-based sprint and jump timing and chest-strap-paired heart-rate variability monitoring. However, the field remains organized around demonstrating measurement accuracy rather than showing that app-guided decisions improve athlete outcomes. A successful pathway for mHealth app development should progress from technical validity, through measurement reliability and responsiveness, to decision rules, and then to practitioner adoption by coaches and athletes, ultimately yielding better athlete outcomes. Full article

Open-channel hydraulic analysis often requires repeated solution of implicit nonlinear equations and numerical integration of gradually varied flow (GVF), which can become computationally demanding in inverse, optimization, and sensitivity applications. This study develops a hydraulically informed artificial neural network (ANN) surrogate framework comprising ten independently trained models for normal and critical depths, alternative and conjugate depths, GVF-related water-surface behavior, and profile-based discharge inference. Hydraulic information is introduced through physically meaningful, and where appropriate dimensionless, variables and reference solutions derived from established governing equations or numerical hydraulic models, while ANN optimization remains data driven. Equation-generated test sets quantified surrogate fidelity, whereas HEC-RAS comparisons were treated as numerical hydraulic cross-verification rather than independent physical validation. The forward surrogates reproduced their reference mappings with high accuracy within the represented domains. Benchmarking against Random Forest, support vector regression, and Gaussian Process Regression for Models 1, 5, and 7 showed no universal algorithmic superiority; however, ANN provided a favorable trade-off among accuracy, relative-error robustness, compactness, and repeated-inference efficiency. For Model 7, ANN inference was approximately 249 times faster than conventional GVF calculation, with development cost recovered after about 1.03 × 10⁵ evaluations. Model 9 inferred discharge with a 4.75% error in the profile-based test. Observation-based assessment using 16 historical Missouri River stage–discharge events showed that direct HEC-RAS inversion yielded a MAPE of 44.71%, whereas observation-only ANN and hybrid HEC-RAS-ANN discrepancy correction reduced MAPE to 10.25% and 9.83%, respectively. The framework is therefore a computational complement to established hydraulic equations and numerical models, with broader field validation and explicit uncertainty treatment required for general deployment. Full article

Background/Objectives: This study aimed to develop machine learning models—specifically logistic regression (LR), random forest (RF), and deep neural network (DNN) models—using initial and 1-month post-stroke clinical data to predict 6-month upper and lower extremity motor functional outcomes in patients with ischemic stroke. Additionally, we sought to evaluate the potential improvement in discriminative performance and clinical utility achieved by integrating 1-month reassessment data. Methods: We analyzed retrospective cohort data from 353 patients with ischemic stroke. Two prediction models were constructed: (1) Model 1, which used only early-stage clinical data, and (2) Model 2, which incorporated both early-stage and 1-month post-stroke clinical data. Model performance and clinical utility were evaluated using the area under the receiver operating characteristic curve (ROC-AUC), DeLong’s test, calibration analysis, decision curve analysis (DCA), and variable importance analysis. Results: Although Model 2, which incorporated 1-month data, generally showed an upward trend in discriminative performance across all models for both upper and lower extremity prediction compared to Model 1, a statistically significant improvement was observed only in the LR model for upper extremity prediction (test AUC increased from 0.889 to 0.990; ΔAUC = +0.102, p = 0.037). For all other models—including the RF and DNN models for the upper extremity, as well as all lower extremity prediction models—the observed increases in AUC did not reach statistical significance according to DeLong’s test. In calibration analyses, the LR model exhibited the most stable calibration for both extremities. In DCA, Model 2 generally yielded a higher net benefit across most threshold probability ranges compared to Model 1 than Model 1 across most threshold probability ranges. Variable importance analysis indicated a shift in the primary contributing variables from initial motor evoked potential parameters in Model 1 to 1-month clinical functional measures in Model 2. Conclusions: Models integrating 1-month reassessment data showed a tendency toward improved discriminative performance compared to those relying solely on initial data. However, as this study was based on a limited sample from a single institution and instability was observed in certain models, external validation using larger, multicenter cohorts is necessary before generalizing these findings. Full article

Early identification of individuals with abnormal glucose metabolism is essential for timely intervention and prevention of diabetes-related complications. Routine laboratory testing generates large amounts of clinical data that may support automated glycemic classification through machine learning approaches. This study aimed to develop and evaluate a machine learning framework for the classification of HbA1c-defined glycemic status using routinely available clinical laboratory features. A retrospective dataset of 1434 individuals with available glycemic measurements was analyzed. Participants were categorized into HbA1c-defined normoglycemic, prediabetic-range, or diabetic-range groups. Three concurrent classification tasks were examined: HbA1c-defined dysglycemia classification, diabetic-range HbA1c classification, and multiclass HbA1c-defined glycemic-status classification. Demographic, biochemical, and hematological variables were used as predictors. Data preprocessing included missing-value handling, feature filtering, and outlier treatment. Several supervised learning algorithms were evaluated, including Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine, and Multinomial Logistic Regression. Model performance was assessed using train–test validation and cross-validation with accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve. For dysglycemia, Gradient Boosting achieved the highest AUC (0.848), while Random Forest achieved the highest accuracy (0.801) and sensitivity (0.908). For diabetic-range HbA1c, Random Forest achieved the highest AUC (0.864), whereas SVM achieved the highest accuracy (0.794). In multiclass classification, Random Forest achieved the highest accuracy (0.610), while Gradient Boosting achieved the highest macro-AUC (0.796) and macro-F1 score (0.603). Pairwise comparisons showed no statistically significant superiority of any classifier after Holm correction. Clinical-baseline and ablation analyses demonstrated that fasting glucose accounted for a substantial proportion of discrimination, with only modest incremental value from additional laboratory variables. These findings support cautious interpretation of routine laboratory-based classification models pending further validation and clinical-utility assessment. Full article

We apply the free energy principle to free markets by treating the Market as a random dynamical system with an attracting set, i.e., some characteristic states. This licenses a normal form for stochastic dynamics that inherits from the Helmholtz–Hodge decomposition. Equipped with this functional form—and a suitable parameterization—one can create a generative model of fluctuations in the value of assets and accompanying indicator variables. This affords the opportunity for prospective (ex ante) prediction, scenario modelling and forecasting that could, in principle, be applied to any complex dynamical system exhibiting stochastic chaos. Here, we illustrate the application to portfolio management—in the context of financial services—and use the (posterior) predictive densities over future paths to evaluate the expected free energy that underwrites active inference. In this application, active inference reduces to risk-sensitive control, which can be used to model the optimal decision-making of an agent or investor. In this setting, an investor is characterized by their prior preferences for a high rate of return under drawdown constraints. Using numerical studies and historical financial data, we quantify the improvement in portfolio management, relative to baseline policies. Full article

Background/Objectives: Human physiological evidence for functional clothing is limited, and garment changes may reflect ordinary material or measurement effects. We described surface-temperature and heart rate variability (HRV) observations before and after wearing black-silica-containing clothing and quantified paired changes and between-participant dispersion. Methods: Ten adults enrolled as healthy volunteers completed this single-center, non-randomized, unblinded, uncontrolled, fixed-order, single-group before–after pilot protocol. No physically matched control textile was used. No directional hypothesis or single primary outcome was prospectively specified. Exploratory domains comprised abdominal and leg surface temperature and eight RR interval (RRI)-derived HRV indices. All participants were analyzed; a post hoc n = 9 quality-control sensitivity analysis excluded one participant with a short post-wearing RRI segment. Effect estimates, 95% confidence intervals (CIs), and Holm-adjusted p values were reported. Between-participant dispersion was secondary and exploratory. Results: Leg surface temperature showed a modest increase of 0.668 °C (95% CI −0.001 to 1.336; raw p = 0.050; Holm p = 0.100); abdominal temperature changed by 0.007 °C (95% CI −0.447 to 0.462). No paired HRV outcome retained support after correction (all Holm p ≥ 0.797). In the secondary dispersion analysis, total power had an after/before log-scale SD ratio of 0.612 (bootstrap 95% CI 0.347 to 0.861; Holm p = 0.031); no dispersion outcome retained support in the n = 9 sensitivity analysis. Conclusions: This small uncontrolled pilot provides hypothesis-generating observations but cannot isolate an effect attributable specifically to black silica or demonstrate autonomic benefit, therapeutic action, or product efficacy. Confirmation requires an adequately powered, randomized, participant-blinded crossover study using physically matched garments and standardized measurement conditions. Full article

In the Ramis River basin of the Peruvian Andes, flood events have become increasingly frequent and intense due to climate variability. However, the basin has limited hydro-meteorological observation records, and its flood generation mechanisms are extremely complex. This situation not only hinders the accurate identification of flood-prone areas, but also limits the effective implementation of flood risk management measures. This study sets three core objectives: to assess flood sensitivity across the basin, identify the dominant factors that influence flood sensitivity, and verify the flood detection performance of multispectral indices. The study adopts two core methods. First, a multi-criteria framework that integrates the Analytic Hierarchy Process (AHP) and Geographic Information System (GIS) is used, incorporating seven flood-related environmental factors and one precipitation triggering variable. Second, the performance of four spectral indices—NDVI, NDWI, SAVI, and MSAVI2 is verified through Spearman correlation analysis, Moran’s I index, and the random forest algorithm. The study finds that landform and geology are the core factors controlling flood sensitivity, with weights of 0.35 and 0.23, respectively. Moderately flood-sensitive areas account for the largest share of the basin, reaching 66% and covering 236.54 km². The flood extent estimated by the spectral indices ranges from 36.73 km² to 101.87 km². Among these indices, NDVI has the strongest spatial correlation with flood-prone areas. The random forest model used in this study has an AUC of 0.6935 and an overall accuracy of 63.51%. The analytical framework proposed in this study is applicable to data-scarce Andean River basins, and the combined use of multispectral indices can provide support for flood risk management and decision-making in this region. Full article

Background: Attention-Deficit/Hyperactivity Disorder (ADHD) affects an estimated 6% of adults in the United States and contributes to a significant economic burden. Telehealth has emerged as a vital tool in the management of ADHD, offering improved access to care, especially for individuals in underserved communities. Despite its growing role, there remain critical gaps in understanding how social determinants of health (SDOH) are associated with disparities in telehealth utilization for ADHD treatment. Objectives and Methods: This study analyzed data from the National Center for Health Statistics (NCHS) Rapid Surveys System (RSS) Round 2: ADHD (October–November 2023), a nationally fielded survey of U.S. adults. Respondents were classified into three groups: never diagnosed, previously diagnosed, and currently diagnosed with ADHD. The study aimed to (1) compare the distribution of SDOH across ADHD status groups and the general adult population to identify factors associated with ADHD diagnosis; (2) assess the homogeneity of SDOH distributions across ADHD groups; (3) evaluate telehealth utilization among adults currently diagnosed with ADHD; and (4) examine the relationship between SDOH and telehealth use for ADHD treatment. Multivariable logistic regression (MVLR) served as a benchmark model, while machine learning (ML) models—including regularized linear regression, support vector machine (SVM), random forest (RF), LightGBM, multilayer perceptron (MLP), and Few-Shot Learning (FSL)—were trained to identify key predictors. Results: A total of 7009 survey responses were analyzed: 124 had a past diagnosis, 444 were currently diagnosed, and the remainder had never been diagnosed with ADHD, corresponding to a current ADHD prevalence of 6.3%. Adults with current ADHD were more likely to be male, single, younger, white, non-homeowners, and frequent users of online health resources. They also reported lower education, income, and financial security. About 70% used telehealth for counseling and prescriptions; insurance covered telehealth visits for 82.32% of users, yet 38.76% reported no coverage of ADHD-related diagnostic or treatment costs. Nineteen SDOH elements across four domains—demographic, socioeconomic, neighborhood/built environment, and healthcare access—were identified as predictors. ML models outperformed MVLR, with SVM and FSL achieving the highest F1 (both 0.63), and FSL the highest recall (0.69). Age, race, marital status, difficulty paying bills, home ownership, education, and household size were the most consistently important variables. Limitations: This study is limited by a cross-sectional design, reliance on self-reported ADHD diagnoses, and a lack of genetic or family-history measures. Additionally, the omission of complex sampling weights limits the national representativeness of these findings. Finally, the small effective sample size poses risks of model overfitting, and the generalizability of the models could not be externally validated due to the unavailability of comparable independent datasets. Conclusions: Despite widespread internet access, disparities in telehealth use for ADHD persist. Among 19 SDOH predictors, age (aOR = 0.56), difficulty paying medical bills (aOR = 2.52), and race (aOR = 1.37) were significantly associated with telehealth use, and all ML models outperformed the MVLR benchmark, though bootstrap CIs overlapped. Future research should incorporate inclusive data collection and stratified modeling to better represent disadvantaged populations and inform equitable access strategies. Full article

Background/Objectives: We investigated whether ingestion of caffeine (~1 h before) was beneficial to subsequent morning (~07:30 h) mood, reactive agility, and cognitive measures in naïve-mild habitual caffeine consumers. Methods: A total of 150 recreationally active males were recruited, of whom 51 participated and completed six sessions as follows: (i) three familiarization sessions of all procedures at 12:00 h, and (ii) three experimental conditions. Participants completed caffeine and placebo trials under double-blind conditions, together with an open-label no-pill control trial using the stratified block randomization method based on baseline performance, which were counterbalanced in a crossover design into either caffeine (CAFF, 300 mg or 2.5–5.3 mg·kg⁻¹ body weight), placebo (PLAC), or no-pill control (NoPill), both ingested at ~06:30 h. For each experimental session, on arrival at the laboratory, questions on sleep, mood states, and caffeine withdrawal were asked. A battery of cognitive performance tests was then administered (trail-making test, Rey’s auditory verbal learning test, and Stroop word–colour interference test). At 30 min of rest and after a 5 min warm-up on a treadmill with associated stretches, rectal and mean skin temperatures (T_r and T_sk) were measured. Thereafter, two 90 s reactive agility tests (RATs) were undertaken, with average movement time(s) and contact number per light recorded. Data were analyzed using a general linear model with repeated measures, but the RAT measures were run as ANCOVA with mass used as a covariate. Results: The battery of cognitive tests showed no significant main effects for condition, whether the body mass of the participants was considered or not. There were no significant main effects of the experimental condition for RAT variables. There was a main effect for “light position”, for mean movement time values (p < 0.001, η²_p = 0.508), and cognitive–motor efficiency (contacts·s⁻¹, p < 0.001, η²_p = 0.582), where pairwise analysis showed mean movement times as higher, and cognitive–motor efficiency values indicating a lower rate of successful responses during movement for lights 1 and 5 (the furthest from the start point corresponding to light 3), rather than lights 2, 3 and 4. There were no significant interactions for condition and light for any of the variables. Conclusions: Early morning ingestion of 300 mg of caffeine had no effect on the battery of cognitive or reactive agility test measures in the population of naïve-mild habitual caffeine consumers. Caffeine ingestion had no effect on T_r, T_sk, mood, or caffeine withdrawal symptom scores, as well as tiredness or alertness, with or without consideration for body mass. Full article

Degraded tax forms and other legal–administrative documents require restoration methods that impose strict constraints on content fidelity. This paper presents a structure-aware restoration framework for degraded tax documents that combines geometric normalization, canonical template guidance, and supervised image restoration within a common alignment space. Documents are first mapped to a canonical layout through homography estimation based on Scale-Invariant Feature Transform (SIFT) and Random Sample Consensus (RANSAC), using the canonical visual template as the reference. Restoration is then performed using template priors and masked constraints designed to preserve the fixed document structure while recovering variable content. Under a fixed-budget comparative protocol, U-Net with template priors achieved the highest visual and structural quality, whereas the proposed hybrid model obtained the best functional OCR performance on synthetic data, reaching a Character Error Rate (CER) of 0.1091 and a Word Error Rate (WER) of 0.3495. As a complementary evaluation on 37 real-world documents with human-generated textual ground truth, restoration increased full-page word coverage across all three OCR engines evaluated, yielding absolute improvements ranging from 4.15 to 18.05 percentage points. Full article

Insect-proof nets provide a non-chemical alternative within integrated pest management (IPM), but their effectiveness against thrips may depend on crop species, season, mesh size, and treatment-associated microclimate conditions. We compared an uncovered and unframed control (CK) and 40-, 60-, and 80-mesh insect-proof nets on thrip abundance in eggplant and potato across two planting seasons in a single-span plastic greenhouse. Weekly plot-level thrip surveys were conducted using a five-point sampling method. Thrips were not identified to species, and the response variable therefore represented total thrip abundance. Treatment effects were estimated with negative binomial (NB1) generalized linear mixed models incorporating a random intercept for plot and an offset for sampling effort; microclimate variables were analyzed with linear mixed models, and exploratory analyses additionally adjusted the NB1 model for measured microclimate covariates to examine whether accounting for these variables altered the estimated net-cage treatment associations. Net-cage treatments generally reduced thrip abundance relative to the control, but the magnitude and consistency of suppression varied markedly with crop and season. The 60-mesh net showed the most consistent direction of suppression, with estimated reductions ranging from 36.0% to 44.4% across the four crop–season contexts; this suppression was statistically supported in both eggplant contexts, but only marginal in the two potato contexts, and this directional consistency did not imply universal superiority in all contexts. The 80-mesh net produced stronger suppression in some cases but was inconsistent, with no detectable suppression in eggplant during the second season. Treatment-related differences in 7-day temperature and relative humidity were generally statistically uncertain, whereas illuminance showed some treatment-associated variation. Adjustment for the measured microclimate covariates did not consistently attenuate the estimated treatment associations, indicating that the measured variables did not provide a simple statistical explanation for the observed context-dependent pattern. Overall, the 60-mesh treatment showed the most consistent estimated suppressive direction under the tested conditions, whereas increasing nominal mesh count did not produce a simple monotonic increase in suppression. The microclimate-adjusted analysis was exploratory and did not provide a causal decomposition of net-cage effects. Full article

Fine-scale mapping of forest types is a prerequisite for accurately assessing forest biomass, biodiversity, ecosystem service values, and carbon budgets. Shihe County, located within an ecotone transitioning from subtropical to temperate monsoon climates in China, was selected as the case study area. By integrating Sentinel-1 SAR and Sentinel-2 multispectral imagery, along with derived vegetation indices, texture features, and backscattering coefficients, as well as statistical features extracted from the 2022 NDVI time-series, we employed a hierarchical classification framework and a random forest algorithm to generate the forest type map. The results indicated the following: (1) With a single-date multi-feature dataset, late winter (3 March) was determined to be the optimal period for forest type classification in Shihe County. (2) Shortwave infrared bands, red-edge bands, the modified vegetation index, the normalized difference red-edge index, mean texture features, and VH-polarization data were the most influential variables. (3) Incorporating yearly NDVI time-series statistical features into the single-date winter subset significantly improved classification performance, yielding an overall accuracy of 86.39% and a Kappa of 0.781, which represents a 10.83% improvement over the baseline. Persistent misclassification between bamboo forests and tea plantations remained a primary constraint on further accuracy enhancement, while the classification accuracy for evergreen broadleaf forest and deciduous coniferous forest exhibited considerable uncertainty, likely attributable to limited reference sample sizes. (4) Deciduous broadleaf forests constituted the dominant land cover type in Shihe County, whereas evergreen coniferous forests, bamboo, and tea plantations were also widely distributed, collectively reflecting the region’s transitional ecological characteristics. Full article

High-dimensional environmental covariates are increasingly available for digital soil mapping (DSM), but their effective use depends on both the feature-selection strategy and the predictive model architecture. However, systematic evidence remains limited regarding how different metaheuristic feature-selection methods interact with standalone and hybrid learning models across multiple soil and groundwater prediction tasks. This study systematically evaluated the interactions between 10 metaheuristic feature-selection algorithms and 13 predictive models, including random forest (RF), convolutional neural network (CNN), recurrent architectures, CNN–recurrent neural network (RNN) hybrids, squeeze-and-excitation (SE)-enhanced hybrids, and iTransformer-based hybrids, across four prediction tasks involving soil organic carbon (SOC), soil–water extract electrical conductivity (ECe), apparent electrical conductivity (ECa), and groundwater level (GWL) in Xinjiang, China. A total of 149 candidate environmental covariates were considered for ECe, SOC, and ECa, whereas 122 candidate covariates were considered for GWL. The results showed that no single feature-selection method consistently performed best across all four targets; instead, predictive performance depended on the interaction among the feature-selection strategy, predictive architecture, and target variable. CNN–RNN hybrid architectures generally achieved higher predictive performance than standalone models, although their benefits varied among prediction targets. The best-performing combinations yielded coefficient of determination (R²) values of 0.9826, 0.6981, 0.8429, and 0.8085 for GWL, SOC, ECe, and ECa, respectively. These findings indicate that target-specific compatibility, rather than aggressive dimensionality reduction or a universally superior algorithm, is a key determinant of predictive performance in high-dimensional DSM. By demonstrating that feature-selection effectiveness is jointly influenced by model architecture and target characteristics, this study provides a methodological reference for developing target-specific digital soil mapping models in arid regions. Full article

The Loess Plateau in China is highly susceptible to frequent landslides and other geological disasters, which have led to substantial losses of natural and human resources and are frequently reported in the news media. Yuzhong County, located east of Lanzhou City, is a mountainous region with considerable development potential. On 7 August 2025, this area experienced a large-scale geological disaster characterized by a compound event involving both landslides and debris flows, resulting in nearly several hundred casualties. With the ongoing urban expansion of Yuzhong County in recent years, the prediction and prevention of geological disasters have become increasingly critical. This study employed three machine learning algorithms—Multiple Logistic Regression (LR), Random Forest (RF), and XGBoost (XG)—to assess landslide susceptibility in Yuzhong County. A total of 169 historical landslide points, supplemented by additional sites identified through field investigations, were compiled, along with 200 non-landslide locations. Multiple environmental factors were incorporated into the models to analyze landslide susceptibility across different areas. Because LR can effectively capture the generalized influence of precipitation variability, it was selected as the primary model for the final susceptibility mapping. To more accurately evaluate the impact of precipitation on landslide occurrence, average seasonal precipitation across the four seasons was used as a predictive factor. To refine the risk assessment at the township level, both raster-based and landslide-unit-based evaluation approaches were adopted. Overlay analyses were then performed by integrating urban infrastructure, population distribution, and predicted landslide hazard zones, while also accounting for the potential influence of extreme precipitation events. The results reveal that the mountainous areas in eastern Mapo Township, southern Xiaokangying Township, southern Xiaguanying Town, and the south-central part of Qingshuiyi Township are high-risk zones prone to group-occurrence landslide disasters. Full article