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/).

The mechanical properties of soil located at cold areas may be deteriorated under freeze-thaw cycle condition. One-part geopolymer (OPG) is a kind of alkaline-activated material by using industrial by-products and solid alkali. Obviously, OPG can replace ordinary portland cement (OPC) as a soil stabilizer in ground improvement, which presents environmental and low-carbon benefits. The assessment of unconfined compressive strength (UCS) is vital for evaluating OPGstabilized soil durability under freeze-thaw conditions, typically demanding extensive resources. Leveraging artificial intelligence, a predictive model can be developed for this purpose. This study collected a small sample size of 216 data points of the OPG-stabilized soil's freeze-thaw behaviour. Three deep learning (DL) models, Backpropagation Neural Network [BPNN], Convolutional Neural Network [CNN], Gated Recurrent Unit [GRU], were trained on the small dataset to predict freeze-thaw performance efficiently, offering a promising approach to streamline assessment processes. In the DL models, the ratio of fly ash (FA) and ground granulated blast furnace slag (GGBFS), freezing temperature and freeze-thaw cycle were taken as the input variables, and the target output was the UCS of the OPG-stabilized soil. Among all the models, the CNN achieved the highest prediction accuracy with R2 of 0.9966, and followed by the BPNN (R2=0.9893) and the GRU (R2=0.9872). After that, the interpretable machine learning methods (i.e., Shapley Additive Explanation [SHAP] and Partial Dependence Plot [PDP]) were utilized for the developed CNN model to further understand the impact of input variables on the outcome predictions. In addition, the morphological analysis was used to verify the freeze-thaw mechanism of the OPG-stabilized soil derived from the interpretable CNN model. It is revealed that the inclusion of FA in the OPG crucially enhanced the freeze-thaw resistance of the OPG-stabilized soil. However, beyond a certain threshold, the addition of FA negatively impacted the freezethaw resistance of OPG-stabilized soil. Freezing temperature was pinpointed as the key factor affecting the properties of the stabilized soil.

Despite the widespread application of data-centric techniques in Geotechnical Engineering, there is a rising need for building trust in the artificial intelligence (AI)-driven safety assessment of road embankments due to its so-called black-boxnature. In addition, from the lens of limit equilibrium approaches, e.g., Bishop, Fellenius, Janbu and Morgenstern-Price, and finite element method, it is essential to carefully examine the interplay of both topological and physical/mechanical properties during the safety factor (FoS) predictions. First, aside from having conventional geotechnical inputs for soil in core and foundation and the height of embankments, this paper codifies geometric features innovatively. The number of slope types with different ratios including 1:1, 1.5:1 and 2:1 as well as the number of berms is introduced. Second, a pool of 19 machine learning (ML) techniques is effortlessly trained on the dataset using an automated ML (AutoML) pipeline to identify the most optimized ML algorithm. Finally, to achieve post-hoc interpretability for the internal mechanism of the input- output relationship unbiasedly, a game-theory-based explainable AI (XAI) method called Shapley additive explanations (SHAP) values is applied. SHAP-aided importance analysis provides human-interpretable insights and indicates height, California bearing ratio, slope type 2:1 and cohesion as the most influential parameters. Exclusively, analyzing hazardous embankments by classifying main and joint contributors exhibits a complex and highly variable influence on the FoS. This paper harnesses the power of XAI tools to enhance reliability and transparency for the rapid FoS prediction of slopes. It targets geotechnical researchers, practitioners, decision-makers, and the general public for the first time.

This study focused on exploring the utilization of a one-part geopolymer (OPG) as a sustainable alternative binder to ordinary Portland cement (OPC) in soil stabilization, offering significant environmental advantages. The unconfined compressive strength (UCS) was the key index for evaluating the efficacy of OPG in soil stabilization, traditionally demanding substantial resources in terms of cost and time. In this research, four distinct deep learning (DL) models (Artificial Neural Network [ANN], Backpropagation Neural Network [BPNN], Convolutional Neural Network [CNN], and Long Short-Term Memory [LSTM]) were employed to predict the UCS of OPG-stabilized soft clay, providing a more efficient and precise methodology. Among these models, CNN exhibited the highest performance (MAE = 0.022, R2 = 0.9938), followed by LSTM (MAE = 0.0274, R2 = 0.9924) and BPNN (MAE = 0.0272, R2 = 0.9921). The Wasserstein Generative Adversarial Network (WGAN) was further utilized to generate additional synthetic samples for expanding the training dataset. The incorporation of the synthetic samples generated by WGAN models into the training set for the DL models led to improved performance. When the number of synthetic samples achieved 200, the WGAN-CNN model provided the most accurate results, with an R2 value of 0.9978 and MAE value of 0.9978. Furthermore, to assess the reliability of the DL models and gain insights into the influence of input variables on the predicted outcomes, interpretable Machine Learning techniques, including a sensitivity analysis, Shapley Additive Explanation (SHAP), and 1D Partial Dependence Plot (PDP) were employed for analyzing and interpreting the CNN and WGAN-CNN models. This research illuminates new aspects of the application of DL models with training on real and synthetic data in evaluating the strength properties of the OPG-stabilized soil, contributing to saving time and cost.