A Novel Approach to Automated Mapping Sub-Weekly Calving Front of Petermann Glacier (2016–2023) Using Sentinel-2 Satellite Data and Segment Anything Model (SAM)

Monitoring glacier calving fronts is critical for understanding ice dynamics and their response to climate change. However, existing methods of mapping ice calving front face limitations in temporal resolution and positional accuracy, hindering effective characterizations of rapid calving processes. To address this, we present an innovative automated workflow for calving front extraction from Sentinel-2 satellite images that integrates the Segment Anything Model (SAM) with Google Earth Engine (GEE) through samgeo and geemap modules. Our approach significantly reduces computational costs and improves efficiency and model generalizability ability. Furthermore, it eliminates the requirements of data preprocessing, manual annotation and model training compared with traditional methods. Applied to Petermann Glacier, we acquire a hightemporal resolution (sub-weekly) and highly accurate datasets of the calving front between 2016-2023. The accuracy assessment of SAM-derived ice front dataset demonstrates a median deviation of within 2.5 pixels (25 m), outperforming the CALFIN's deep learning-based results approximately 3 pixels (∼90m). Validation demonstrates strong adaptability and robustness with tuning points prompts. This automated workflow provides a precise and efficient solution for monitoring glacier advance and retreat, enabling improved quantification of ice sheet mass loss and enhanced assessment of glacier-related hazards.