AI-Driven Approaches applied on Time-Lapse Imagery to Monitor Landform Kinematics

Moving landforms, such as active rock glaciers and landslides, can pose significant hazards, particularly in densely populated regions such as the European Alps. Traditional techniques used to monitor landform kinematics, including in-situ differential Global Navigation Satellite System (GNSS) and georeferenced Total Station (TS) measurements, face limitations in capturing the rapid and localized movements due to environmental constraints and restricted spatial coverage. Remote sensing methods provide improved spatial resolution but often fall short in temporal resolution, limiting their ability to capture sub-seasonal dynamics.

This study presents a novel methodology that integrates Artificial Intelligence (AI) and monoscopic time-lapse imaging to address these challenges, enabling high-temporal-resolution velocity estimation for dynamic landform processes. Focusing on the Grabengufer site in the Swiss Alps, we applied our approach to time-lapse datasets capturing a fast-moving landslide and rock glacier. Key innovations include the Persistent Independent Particle tracking (PIPs++, Zheng et al., 2023) model for 2D image-based point tracking and a robust image-to-geometry registration process that transfers 2D measurements into 3D object space, facilitating velocity analysis. These processes are supported by GIRAFFE, an AI-based tool utilizing the LightGlue matching algorithm for precise feature registration.

Our methodology was validated against GNSS and TS surveys, demonstrating its ability to deliver spatially comprehensive and temporally detailed velocity data. The results revealed previously unattainable spatio-temporal patterns of landform activity, highlighting the suitability of this approach for monitoring rapid and localized changes. By leveraging existing time-lapse imagery, the methodology provides a low-cost alternative to traditional techniques, with potential applications in less-developed regions where resources for monitoring are limited.

This research underscores the potential of integrating time-lapse images, AI, and geomorphometric analysis to enhance the understanding of landslide behaviour and related hazards. The proposed approach not only advances the capabilities of landslide monitoring but also provides actionable data for long- and short-term risk reduction. Its versatility and cost-effectiveness make it a valuable tool for addressing landslide risks worldwide, contributing to more effective hazard assessment, climate change adaptation, and infrastructure safety planning.

 

Zheng, Y., Harley, A. W., Shen, B., Wetzstein, G., & Guibas, L. J. (2023). Pointodyssey: A large-scale synthetic dataset for long-term point tracking. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 19855-19865).

Hendrickx, H., Elias, M., Blanch, X., Delaloye, R., and Eltner, A.: AI-Based Tracking of Fast-Moving Alpine Landforms Using High Frequency Monoscopic Time-Lapse Imagery, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2024-2570, 2024.