Fusing a foundation model with spatiotemporal correction for automated long-term tracking of rift propagation from satellite imagery: a case study of the amery ice shelf

Long-term satellite monitoring of rift propagation is essential for assessing Antarctic ice shelf stability and projecting sea-level rise. Currently, long-term tracking of rifts still relies on time-consuming manual interpretation, significantly affecting its feasibility for large-scale rift monitoring. Here, we present a novel framework for automatically tracking rift propagation, which fuses the fine-tuned visual foundation model (Segment Anything Model, SAM) and spatiotemporal correction. The SAM enables automated rift extraction using only a few samples without any manual prompts, while the spatiotemporal correction improves the precision of rift mapping for geometric measurements. We applied this framework to automatically monitor key rifts on the Amery Ice Shelf using 768 Landsat 8/9 and Sentinel-2 images from 2015 to 2025, achieving an F1 score of 98.6% across 11 independent scenes. Rift length and width measurements were in agreement with manual interpretation, with R2 values of 0.979 and 0.999, respectively. Furthermore, our extended monitoring revealed recent structural changes within the rift system on the Amery Ice Shelf. This study offers a new way for automated rift tracking with the foundation model for the first time, having significant potential for long-term monitoring of pan-Antarctic rifts.