To address the challenge of the complex and extensive seismic design elements of tunnels, which are difficult to be accurately described using mathematical functions, a novel model combining convolutional neural networks (CNN), gated recurrent units (GRU), and an attention mechanism is proposed. Firstly, based on actual engineering examples, the tunnel dimensions and site soil information are determined to establish a numerical model of tunnel seismic response and verify its reliability. Then, the soil parameters, seismic motion amplitude, tunnel depth, and overlying water depth are selected for systematic analysis of the displacement momentum (DM) and time of maximum damage occurrence (TMDO). The parameters with higher influence are chosen as input variables, while the calculated DM and TMDO from the reliable numerical model are selected as the output variables to be predicted. Next, integrating the GRU model to capture long-term dependencies in time series, the CNN model to extract spatial features, and the attention mechanism to handle complex relationships among multiple variables, the CNN-GRU-Attention prediction model was established. By generating dataset samples through numerical simulation, accurate predictions of DM and TMDO were achieved. Finally, using the proposed model to establish the objective function relationship between input and output parameters, employing the NonDominated Sorting Genetic Algorithm II (NSGA-II) to find the optimal input design features, achieving the optimal design of tunnel seismic performance. The results show that: (1) The calculation results of the numerical model for tunnel seismic response conform to general research findings, indicating sufficient reliability. (2) The error compensation and dynamic updating mechanisms improved prediction accuracy. The R2 values for the training set reach 0.973 and 0.982 respectively. (3) Optimizing DM and TMDO using the NSGA-II algorithm leads to a 23.42% reduction in DM and a 18.71% increase in TMDO. After optimization, tunnel displacement is reduced, damage is delayed, and seismic performance is significantly improved.

Solar panels are essential for converting sunlight into electricity. Still, environmental factors can significantly compromise their efficiency and performance, particularly the accumulation of soiling on their surfaces or damage. This study proposes a hybrid model comprising an ensemble of deep-learning models to distinguish between soiled and damaged solar panels and their corresponding conditions. Our approach utilizes pretrained deep learning models, fine-tuned for detecting soiling or damage on photovoltaic (PV) panels, to extract relevant features and build efficient classifiers. Introducing a post-processing ensemble model improves classification metrics compared to a single deep-learning model. Combining Convolutional Neural Networks and Vision Transformers in an ensemble model achieves the highest accuracy, with 96.3% for damage and soiling detection and 91.8% for damage and soiling type identification. These results significantly outperform one-tier deep learning models, which attain an accuracy of 87.7% when classifying all possible damage and soiling categories.

Floods and erosion are natural hazards that present a substantial risks to both human and ecological systems, particularly in coastal regions. Flooding occurs when water inundates typically dry areas, causing widespread damage, while erosion gradually depletes soil and rock through processes driven by water and wind. This study proposes an innovative approach that integrates Deep Neural Decision Forest (DeepNDF), Feedforward Neural Network (FNN), autoencoders, and Bidirectional Recurrent Neural Networks (Bi-RNN) models for flood prediction, enhanced through transfer learning for erosion mapping in coastal environments. Utilizing multi-source datasets from the United States Geological Survey (USGS), Climate Hazards Group InfraRed Precipitation with Station (CHIRPS), the National Algerian Institute of Cartography, and Sentinel-2 imagery, the key conditioning factors using Geographic Information System (GIS) were generated. The conditioning factors included elevation, slope, flow direction, curvature, distance from rivers, distance from roads, hillshade, topographic wetness index (TWI), stream power index (SPI), geology, and land use/land cover (LULC), as well as rainfall. To ensure the modeling reliability, the performance was rigorously evaluated using multiple statistical metrics, including the Area Under the Curve-Receiver Operating Characteristic (AUC-ROC), Precision, Recall, and F1 Score. The DeepNDF model achieved the highest performance for flood prediction with an AUC-ROC of 0.97, Precision of 0.93, Recall of 0.92, and an F1 Score of 0.925, while the transfer learning approach significantly improved erosion prediction, reaching an AUC-ROC of 0.92, Precision of 0.90, Recall of 0.92, and an F1 Score of 0.91. The analysis indicated that flood risks predominantly affected rangeland (18.68%) and bare ground (20.48%), while cropland was found to face the highest erosion risk, affecting approximately 3,471 km2. This research advances predictive modelling in hydrology and environmental science, providing valuable insights for disaster mitigation and resilience planning strategies in coastal areas.

Shield tunnels in operation are often affected by complex geological conditions, environmental factors, and structural aging, leading to cumulative damage in the segments and, consequently, increased deformation that compromises structural safety. To investigate the deformation behavior of tunnel linings under random damage conditions, this study integrates finite element numerical simulation with deep learning techniques to analyze and predict the deformation of shield tunnel segments. First, a refined three-dimensional finite element model was established, and a random damage modeling method was developed to simulate the deformation evolution of tunnel segments under different damage ratios. Additionally, a statistical analysis was conducted to assess the uncertainty in deformation caused by random damage. Furthermore, this study introduces a convolutional neural network (CNN) surrogate model to enable the rapid prediction of shield tunnel deformation under random damage conditions. The results indicate that as the damage ratio increases, both the mean deformation and its variability progressively rise, leading to increased deformation instability, demonstrating the cumulative effect of damage on segment deformation. Moreover, the 1D-CNN surrogate model was trained using finite element computation results, and predictions on the test dataset showed excellent agreement with FEM calculations. The surrogate model achieved a correlation coefficient (R2) exceeding 0.95 and an RMSE below 0.016 mm, confirming its ability to accurately predict the deformation of tunnel segments across different damage conditions. To the best of our knowledge, the finite-element-deep-learning hybrid approach proposed in this study provides a valuable theoretical foundation for predicting the deformation of in-service shield tunnels and assessing structural safety, offering scientific guidance for tunnel safety evaluation and damage repair strategies.

In this study, a novel data-driven approach is carried out to predict the pore pressure generation of liquefiable clean sands during cyclic loading. An extensive and comprehensive database of actual stress-controlled cyclic simple shear test results in terms of pore pressure time histories is gathered from a large number of experiments. While the classical machine learning (ML) algorithms help predict the number of liquefaction cycles in a few models, the desired level of accuracy in predicting the actual trend and robustness in pore pressure build-up is only achieved in deep learning (DL) methods. Results indicate that the Long-Short Term Memory (LSTM) working model, employed with Stacked LSTM and the Windowing data processing method, is necessary for making fairly good cyclic pore pressure build-up predictions. This study proposes a model that can ultimately be utilised to predict the pore pressure response of in-situ liquefiable sandy soil layers without resorting to plasticity-based complex theoretical models, which has been the current practice. The robustness achieved in the model reassures the reliability of the study, raising confidence in developing data-driven constitutive models for soils that have the potential to replace conventional plasticity-based theories.

Background and AimsSoil salinization is a major cause of land degradation and ecological damage. Traditional soil salinity monitoring techniques are limited in coverage and scalability, while remote sensing offers broader applicability and efficiency. This study addresses spatiotemporal variations in soil salt content (SSC) inversion across crop types in Tongliao City, Inner Mongolia, China, using an innovative integration of multi-temporal data and crop cover types, improving remote sensing monitoring accuracy.MethodsField sampling data and Sentinel-2 images from June to September in 2021 and 2022 were utilized. The deep learning U-net model classified key crops, including sunflower (33%), beet (12%), and maize (55%), and analyzed the effects of crop coverage on SSC across multiple time series. Six spectral variables were selected using the SVR-RFE model (R2 = 0.994, MAE = 0.016). SSC prediction models were developed using three machine learning methods (DBO-RF, PSO-SVM, BO-BP) and a deep learning method (Transformer).ResultsConsidering crop coverage variations improved the sensitivity of spectral variables to SSC response, enhancing predictive accuracy and model stability. Crop classification showed that the salinity index (SIs) correlated more strongly with SSC than the vegetation index (VIs), with SI6 having the highest correlation coefficient of 0.50. The Transformer model, using multi-time series data, outperformed other algorithms, achieving an average R2 of 0.71. The SSC inversion map from the Transformer model closely matched field survey trends.ConclusionThis research provides a novel approach to soil salinity prediction using satellite remote sensing, offering a scalable solution for monitoring salinization and valuable insights for environmental management and agricultural planning.

The outbreak of Pine Shoot Beetle (PSB, Tomicus spp.) posed a significant threat to the health of Yunnan pine forests, necessitating the development of an efficient and accurate remote sensing monitoring method. The integration of unmanned aerial vehicle (UAV) imagery and deep learning algorithms shows great potential for monitoring forest-damaged trees. Previous studies have utilized various deep learning semantic segmentation models for identifying damaged trees in forested areas; however, these approaches were constrained by limited accuracy and misclassification issues, particularly in complex forest backgrounds. This study evaluated the performance of five semantic segmentation models in identifying PSB-damaged trees (UNet, UNet++, PAN, DeepLabV3+ and FPN). Experimental results showed that the FPN model outperformed the others in terms of segmentation precision (0.8341), F1 score (0.8352), IoU (0.7239), mIoU (0.7185) and validation accuracy (0.9687). Under the pure Yunnan pine background, the FPN model demonstrated the best segmentation performance, followed by mixed grassland-Yunnan pine backgrounds. Its performance was the poorest in mixed bare soil-Yunnan pine background. Notably, even under this challenging background, FPN still effectively identified diseased trees, with only a 1.7% reduction in precision compared to the pure Yunnan pine background (0.9892). The proposed method in this study contributed to the rapid and accurate monitoring of PSB-damaged trees, providing valuable technical support for the prevention and management of PSB.

Forecasting landslide deformation is challenging due to influence of various internal and external factors on the occurrence of systemic and localized heterogeneities. Despite the potential to improve landslide predictability, deep learning has yet to be sufficiently explored for complex deformation patterns associated with landslides and is inherently opaque. Herein, we developed a holistic landslide deformation forecasting method that considers spatiotemporal correlations of landslide deformation by integrating domain knowledge into interpretable deep learning. By spatially capturing the interconnections between multiple deformations from different observation points, our method contributes to the understanding and forecasting of landslide systematic behavior. By integrating specific domain knowledge relevant to each observation point and merging internal properties with external variables, the local heterogeneity is considered in our method, identifying deformation temporal patterns in different landslide zones. Case studies involving reservoir-induced landslides and creeping landslides demonstrated that our approach (1) enhances the accuracy of landslide deformation forecasting, (2) identifies significant contributing factors and their influence on spatiotemporal deformation characteristics, and (3) demonstrates how identifying these factors and patterns facilitates landslide forecasting. Our research offers a promising and pragmatic pathway toward a deeper understanding and forecasting of complex landslide behaviors. (c) 2025 Institute of Rock and Soil Mechanics, Chinese Academy of Sciences. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/ 4.0/).

Estimating the landscape and soil freeze-thaw (FT) dynamics in the Northern Hemisphere (NH) is crucial for understanding permafrost response to global warming and changes in regional and global carbon budgets. A new framework for surface FT-cycle retrievals using L-band microwave radiometry based on a deep convolutional autoencoder neural network is presented. This framework defines the landscape FT-cycle retrieval as a time-series anomaly detection problem, considering the frozen states as normal and the thawed states as anomalies. The autoencoder retrieves the FT-cycle probabilistically through supervised reconstruction of the brightness temperature (TB) time series using a contrastive loss function that minimizes (maximizes) the reconstruction error for the peak winter (summer). Using the data provided by the Soil Moisture Active Passive (SMAP) satellite, it is demonstrated that the framework learns to isolate the landscape FT states over different land surface types with varying complexities related to the radiometric characteristics of snow cover, lake-ice phenology, and vegetation canopy. The consistency of the retrievals is assessed over Alaska using in situ observations, demonstrating an 11% improvement in accuracy and reduced uncertainties compared to traditional methods that rely on thresholding the normalized polarization ratio (NPR).

Advanced machine learning and deep Learning modeling applications for landslide susceptibility mapping are becoming increasingly popular. This study applied a deep learning model (DL) with a multilayer neural network to landslide research in the Phuoc Son district, Quang Nam province. Two methods for selecting conditioning factors, Correlation Attribute and OneR, were used to choose 12 condition parameters for landslides (Slope, Relief, Elevation, Distance to road, Rainfall, Land use, Weathering crust, Geology, Aspect, Soil, Distance to fault, and Curvature). Comparing the predicted results with two standard models, Naive Bayes (NB) and Support Vector Machine (SVM), showed that the DL model has higher and better prediction performance. Accordingly, the prediction performance of the DL model on the training dataset was ACC = 92.12%, AUC = 0.970, and on the validation dataset was ACC = 87.52, AUC = 0.944. The LSM developed based on the DL model indicates that areas with high landslide susceptibility are primarily concentrated in the southern part of the study area. These findings could be highly beneficial for urban planning management, risk management, and efforts to prevent and mitigate the damage caused by landslides in Phuoc Son.