AI-based tracking of fast-moving alpine landforms using high-frequency monoscopic time-lapse imagery

Abstract. Active rock glaciers and landslides are dynamic landforms in high mountain environments, where their geomorphic activity can pose significant hazards, especially in densely populated regions such as the European Alps. Moreover, active rock glaciers reflect the long-term thermal state of permafrost and respond sensitively to climate change. Traditional monitoring methods, such as in situ differential Global Navigation Satellite System (GNSS) and georeferenced total station (TS) measurements, face challenges in measuring the rapid movements of these landforms due to environmental constraints and limited spatial coverage. Remote sensing techniques offer improved spatial resolution but often lack the necessary temporal resolution to capture sub-seasonal variations. In this study, we introduce a novel approach utilising monoscopic time-lapse image sequences and artificial intelligence (AI) for high-temporal-resolution velocity estimation, applied to two subsets of time-lapse datasets capturing a fast-moving landslide and rock glacier at the Grabengufer site (Swiss Alps). Specifically, we employed the Persistent Independent Particle tracker (PIPs++) model for 2D image point tracking and the image-to-geometry registration to transfer the measured 2D image points into 3D object space and further into velocity data. For the latter, we use an in-house tool called GIRAFFE, which employs the AI-based LightGlue matching algorithm. This methodology was validated against GNSS and TS surveys, demonstrating its capability to provide spatially and temporally detailed velocity information. Our findings highlight the potential of image-driven methodologies to enhance the understanding of dynamic landform processes, revealing spatiotemporal patterns previously unattainable with conventional monitoring techniques. By leveraging existing time-lapse data, our method offers a cost-effective solution for monitoring various geohazards, from rock glaciers to landslides, with implications for enhancing alpine safety. This study marks the pioneering application of AI-based methodologies in environmental monitoring using time-lapse image data, promising advancements in both research and practical applications within geomorphic studies.