HIGH-FREQUENCY 4D MONITORING OF ALPINE GLACIERS USING LOW-COST TIME-LAPSE CAMERAS AND DEEP LEARNING STRUCTURE-FROM-MOTION

Abstract

Monitoring glaciers in alpine regions over short-term intervals is essential for understanding their response to climate change. This paper presents a low-cost time-lapse camera system designed for continuous 4D glacier monitoring. The system integrates daily 3D reconstruction from stereo cameras and surface velocity estimation from monoscopic camera using Digital Image Correlation (DIC). To overcome challenges posed by wide camera baselines and achieve accurate 3D reconstruction under suboptimal viewing conditions, state-of-the-art deep learning feature matching algorithms are employed in stereo reconstruction, which is not achievable with traditional methods. A pilot study conducted at the debris-covered Belvedere Glacier (Italian Alps) demonstrates the effectiveness of the proposed approach in accurately estimating daily glacier dynamics, including surface kinematics, ice volume loss, and glacier retreat. The combination of stereo 3D reconstruction and DIC reveals a significant short-term correlation between air temperature, glacier surface velocity, and ablation, providing insights into the glacier's response to external factors. The proposed system offers ease of replication and can be readily applied to other study sites.