Three-dimensional Deformation Field Construction Method for Landslides Using ICP Algorithm and Physics-Informed Deep Learning

Deformation monitoring is essential for the early warning of landslide disasters. Constructing a three-dimensional (3D) deformation field can effectively reveal the overall deformation characteristics of a landslide, thereby enhancing monitoring and early warning performance. Using limited discrete monitoring data obtained from experiments, this study proposed a novel method for reconstructing the 3D deformation field of landslides. An improved iterative closest point algorithm was applied to process multi-temporal point cloud data acquired from terrestrial laser scanning in segments, yielding the surface deformation field of the slope. The potential slip depth was initially estimated using a balanced cross-section algorithm based on surface deformation. Subsequently, a physics-informed deep learning model was employed to integrate limited surface and subsurface deformation data, enabling the reconstruction of the 3D deformation field. Compared to traditional numerical simulation methods and geological interpolation techniques, the proposed approach leverages field-measured data for modeling, offering higher accuracy and a simpler workflow, which facilitates effective real-time monitoring of landslide deformation fields. The reconstructed 3D deformation field aids in identifying potential rupture and slip surfaces, predicting deformation at unmonitored locations, and providing clearer and more intuitive observational outcomes.