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/).

Constructing an interpretable model for the long-term deformation Structural Health Monitoring (SHM) of earthrock dams is of great significance for improving the safety state evaluation and monitoring effect. In this paper, a physics -data -driven model for the deformation SHM of earth -rock dams is proposed based on deep mechanism knowledge distillation. Firstly, the deterministic model is established based on the Finite Element Model (FEM) and outputs the hydraulic load component curve and aging component curve. Then a regression prediction model (HTSGAN) between influencing factors and deformation measurements at multiple measurement points is established based on the Graph Convolutional Network (GCN) and attention mechanism. Finally, the TeacherHydraulic -Time -Seepage Graph Attention Networks (T-HTSGAN) model is established based on the featurebased multi -teacher knowledge distillation using the knowledge of hydraulic loading physics and soil -rock creep physics of the FEM for mechanism constraints. The model effectively solves the problems of poor model interpretability and lack of physics knowledge constraints in previous earth -rock dam SHM models. The research results are applied to a project of a 185.5 -meter -high concrete -faced rockfill dam, and the predictive performance of the model is more effective and stable through the comparison of six baseline models. The comparative analysis of the component curves proves the effectiveness of the proposed knowledge distillation method for mechanism constraints and improves the interpretability of the neural network model. Therefore, the model is more suitable for engineering applications.