Aug 2026· Applied Sciences· Vol 16, pp. 7671· 0 citations· 33 references
Abstract
Long-term dam safety assessment relies on continuous monitoring data from multiple spatially distributed measurement points. However, monitoring data are affected by measurement noise, environmental disturbances, and model errors, while the spatial correlation among monitoring points is often ignored, leading to biased uncertainty estimation and unreliable prediction intervals. To address these issues, this study proposes a spatial-correlation-aware distribution-adaptive interval prediction method for dam monitoring via two-level uncertainty fusion. Measurement random noise is first separated using a filtering strategy, and its uncertainty is updated by incorporating the spatial correlation among multiple monitoring points. A regression model is then established based on the filtered monitoring data, and model prediction uncertainty is quantified from the residual distribution. The two uncertainty components are further integrated to construct distribution-adaptive asymmetric prediction intervals. The proposed method is verified using deformation monitoring data from the PB high core rockfill dam. The results show that the proposed method achieves an average PICP of 0.9898 on the training set, close to the target coverage level of 0.99, while reducing the average NMPIW by 16.9% and 13.4% compared with the symmetric interval method and the traditional 3σ method, respectively. On the validation set, the average NMPIW is further reduced by 18.4% and 26.5%, demonstrating that the proposed method can provide more compact and informative prediction intervals for refined dam safety monitoring.
Dam displacement forecasting is important for assessing whether long-term structural responses remain consistent with established operational behavior. In multi-point monitoring systems, however, irregular survey-line layouts and unequal numbers of monitoring points make it difficult to organize long-term records while preserving their engineering meaning. This study develops a field-oriented forecasting framework by reconstructing daily observations into a structured displacement-field object defined by survey-line order, monitoring-point alignment, and three displacement components. A valid-position-aware protocol is introduced to distinguish actual monitoring locations from structural padding, ensuring that model training and evaluation remain restricted to the same physical monitoring definition. Using long-term operational records from the Tianshengqiao First Dam, four representative models, namely SimVP, SimVPv2, PatchTST, and TimesNet, are evaluated under the same chronological split, causal forward-fill-only preprocessing, input window, prediction horizon, and evaluation boundary. All four models achieve strong predictive performance, with R2 values above 0.97 in the X direction and above 0.99 in the Y and Z directions. No single trained model or forecasting route exhibits a consistent advantage across all displacement components and evaluation metrics. Under the present single-dam, case-specific setting, the relative ranking varies with displacement direction and forecasting horizon and should not be interpreted as evidence of general direction-specific suitability for any particular architecture. At the route level, the field-based route retains a slight advantage in Y-direction forecasting and overall MAE, whereas the sequence-based route remains competitive for Z-direction displacement and longer-horizon X-direction prediction. The proposed framework provides a practical and physically consistent digital representation for organizing irregular monitoring records, comparing forecasting routes, and supporting deployment-oriented model selection and subsequent model adaptation.
Xin Xu, Jun Zhang, Shuangping Li et al.· Engineer· 0 citations
This paper addresses the need for multi-source data fusion and risk early warning in slope monitoring by proposing a modeling method that integrates data calibration, feature interpretation, and hierarchical early warning. First, linear regression is used to calibrate fiber-optic displacement data, and data consistency is improved through anomaly removal, multiple interpolation, and wavelet denoising. Subsequently, a LightGBM regression model is constructed to characterize the nonlinear relationship between surface displacement and factors such as rainfall, pore water pressure, microseismic events, and deep-seated displacement. The SHAP method is introduced to explain the contributions of each factor, thereby enhancing the interpretability of the model results. Building on this, Lasso regression is used to screen for key variables and reduce redundant information, followed by the establishment of a probabilistic representation of displacement velocity and graded early warning thresholds based on the Weibull distribution. This method balances prediction accuracy, physical plausibility, and engineering applicability, and can serve as a reference for identifying complex slope deformations, analyzing key controlling factors, and making risk warning decisions.
Monitoring mining infrastructure is essential for ensuring operational safety and minimizing environmental risks associated with subsidence and structural instability. However, existing approaches often rely on single-source data, limiting their ability to capture complex and dynamic deformation processes. This study proposes an integrated multi-source monitoring framework that combines satellite-based Interferometric Synthetic Aperture Radar (InSAR), unmanned aerial vehicle (UAV) observations, ground-based measurements, and advanced digital analysis techniques. The methodology integrates multi-temporal InSAR time-series analysis with high-resolution UAV-derived data within a unified geospatial environment, supported by machine learning models for pattern recognition and prediction. The results reveal clear spatial and temporal deformation patterns in mining areas, with subsidence rates up to −50 mm/year in active extraction zones. Time-series analysis shows that deformation evolves nonlinearly, underscoring the importance of continuous monitoring for early risk detection. Validation against ground-based measurements confirms the reliability of the proposed approach, with root-mean-square error (RMSE) values in the range of 1–2 mm. The integration of multi-source data significantly improves monitoring accuracy and enables detailed analysis across different spatial scales. Furthermore, the application of machine learning techniques enhances predictive capability, allowing identification of potential instability zones before critical failure occurs.
Nurbek Abdixamidov, Maksuda Juraeva, Nafosat Meyliyeva et al.· Geotechnology, Mining and Ra...· 0 citations
Errors in weather forecasts remain a critical challenge for high‐resolution numerical weather prediction (NWP) and artificial intelligence forecast models. Traditional verification approaches, such as Threat Score (TS) and Equitable Threat Score (ETS), suffer from “double penalty” issues due to spatial‐temporal mismatches and lack of spatial distribution fidelity. In this study, we introduce an integrated multi‐component error verification approach (IMEV) that decomposes forecast errors into three physically interpretable components: accumulation error (systematic intensity discrepancy), area error (areal coverage and spatial extent mismatch), and pattern error (distributional dissimilarity and organization difference). Pattern error is evaluated by the similarity of intensity distribution features between the forecast and observation fields. These intensity distribution features are represented using probability density histograms of meteorological elements. Pattern error is quantified using Jensen‐Shannon divergence of probability density distributions, avoiding complex feature matching while capturing spatial pattern discrepancies. Two representative precipitation cases are analyzed to validate this method. Results demonstrate that the proposed method aligns closely with subjective analysis, outperforming traditional metrics like TS and ETS in capturing spatial distribution discrepancies. By prioritizing dominant error dimensions (accumulation, area, and pattern), the method downplays fine‐scale spatial details while preserving essential distributional characteristics through probability density analysis. Applying the IMEV algorithm within small‐sized sliding windows can, to a certain extent, allow for partial sensitivity to spatial displacements. This avoids computationally intensive feature matching, enabling efficient error quantification.
Xiyu Mu, Qi Xu, Guoqing Liu et al.· Journal of Geophysical Resea...· 0 citations
Traditional identification of accident-prone road sections relies on fixed-unit statistics, which has disadvantages that include missing spatial continuity of risk, boundary distortion, and information “averaging,” and struggles to support differentiated safety management. To address this issue, this study proposes a road accident risk field reconstruction model integrating dynamic spatial attenuation and probability-severity dual-dimensional coupling. First, based on the information diffusion principle and Gaussian kernel density estimation (KDE), a framework for converting discrete accident points into a continuous risk probability field is constructed, and a dynamic standard deviation function is proposed. This function enables the Gaussian kernel standard deviation to dynamically adapt to road design speed and cross-sectional type (with/without median divider), clarifying its physical correlation with drivers’ sight distance requirements. Second, by integrating casualties and direct economic losses, a continuous accident severity field is established through an accident equivalent loss function; the two fields are discretized using the quantile method, and combined with a risk matrix to generate comprehensive risk ratings (Levels I–V) via coupling. Verification using accident data from a Class II mountain highway in Guangxi shows the following: (1) the boundaries of the risk field generated by the model fit the road alignment well, with a probability field gradient smoothness of 0.0068, which can mitigate the boundary effect and step effect of the fixed-unit method; (2) compared with KDE with fixed bandwidth (0.35 km) and fixed-length segmentation method (0.5 km), the area under the curve (AUC) of the model’s probability field reaches 0.9641 (increased by 12.79% and 15.31%, respectively), the AUC of the severity field is 0.8669 (increased by 15.24% and 5.88%, respectively), and the Pearson correlation coefficient between the probability field and accident data is 0.4772 (increased by 11.01% and 3.77%, respectively), indicating that the identification accuracy and data fit have been improved; and (3) dual-dimensional coupling can distinguish risk patterns of high-frequency low-loss, low-frequency high-loss, and high-frequency high-loss, providing a quantitative basis for differentiated management. This model promotes the evolution of road risk assessment from discrete statistics to continuous field analysis, and can offer technical support for the optimal allocation of safety resources.
Jianfeng Liu, Fuyuan Luo, Jianqiu Chen et al.· Journal of Transportation En...· 0 citations
Low-cost air quality sensors can expand PM2.5 monitoring networks but require calibration against regulatory-grade monitors to correct systematic bias towards trustworthy observations. Although interest in predictive uncertainty for air quality estimation has increased, the reliability of uncertainty under spatial sparsity and wildfire-induced distribution shift remains poorly understood. With the increasing use of low-cost sensors during wildfire-smoke episodes, uncertainty quantification during calibration is essential to determine when corrected PM2.5 estimates are reliable enough to support near-real-time alert, public-health, and exposure-related decisions. This study presents a spatial uncertainty evaluation framework for trustworthy AI-enabled environmental sensing. The framework is implemented with a transformer-based PM2.5 calibration model and two uncertainty quantification methods, GeoConformal Prediction (GCP) and Monte Carlo Dropout (MCD), using sensor pairs in California and the Northeast United States. The calibration model achieved strong performance at short spatial distances, with test R2 values of 0.89 and 0.91 at the 1 km threshold in California and the Northeast, respectively, with accuracy declining as sensor separation increased. GCP generally produced calibration curves closer to the ideal diagonal, while MCD generated tighter prediction intervals under normal conditions. During wildfire events, the uncertainty performance depended on sensor separation. At short distances, MCD expanded its uncertainty intervals and captured PM2.5 spikes more effectively than GCP (71% vs. 61% coverage). At larger separations, MCD captured only 44% of elevated observations, whereas GCP widened its intervals and achieved 83% coverage. These results demonstrate that uncertainty reliability is strongly influenced by spatial separation and environmental conditions, highlighting the need for uncertainty-aware and spatially adaptive calibration of low-cost PM2.5 sensors for trustworthy actionable air quality observation.
Anusha Srirenganathan Malarvizhi, Kaylee A. Smith, Seren Smith et al.· Atmosphere· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
