Three-Dimensional Characterization of Pan-Antarctic Ice Shelf Fracture: An Integrated Deep Learning and Hydrological Analysis Framework

Abstract

Fractures represent vulnerable discontinuities formed under stress conditions, with their 3-D morphological parameters serving as pivotal indicators for assessing ice shelf dynamic stability. The current fracture monitoring system primarily focuses on 2-D feature analysis, and there is insufficient 3-D systematic monitoring of vertical extension processes. Based on the reference elevation model of Antarctica (REMA) DEM data, this study integrates deep learning semantic segmentation with hydrological terrain analysis methods to construct a framework for extracting fracture depth information. For the first time, a comprehensive dataset of fracture depths across the Antarctic ice shelves is created, and based on this dataset, the 3-D extent of ice shelf damage is quantified and evaluated. The study shows that the average depth of fractures in ice shelves is 8.17 m, with differences between ice shelves reaching up to ten times. Notable spatial variations in fracture depth are also observed within ice shelves. The depth distribution of fractures exhibits significant spatial coupling with the stretching stress field of the ice shelf. The 3-D morphological parameters of the ice shelf (average depth, area density, volume density, and penetration rate) exhibit significant spatial heterogeneity. This study fills the gap in the vertical dimension of fracture 3-D modeling, providing essential data support for ice shelf stability research.