Deep learning the flow law of Antarctic ice shelves

• 成果类型: Article
• 署名作者: Wang, Yongji; Lai, Ching-Yao; Prior, David J.; Cowen-Breen, Charlie
• 署名单位: Stanford University; New York University; Princeton University; University of Otago; Princeton University
• 刊物名称: SCIENCE
• ISSN/ISSBN: 0036-8489
• DOI: 10.1126/science.adp3300
• 发表日期: 2025-03-14
• 页码: 1219-1224
• 关键词: informed neural-networks; fracture-mechanics; sheet dynamics; grain-size; stream-b; part 1; MODEL; assimilation; STABILITY; rheology

摘要

Antarctic ice shelves buttress the grounded ice sheet, mitigating global sea level rise. However, fundamental mechanical properties, such as the ice flow law and viscosity structure, remain under debate. In this work, by leveraging remote-sensing data and physics-informed deep learning, we provide evidence over several ice shelves that the flow law follows a grain size-sensitive composite rheology in the compression zone. In the extension zone, we found that ice exhibits anisotropic properties. We constructed ice shelf-wide anisotropic viscosity maps that capture the suture zones, which inhibit rift propagation. The inferred stress exponent near the grounding zone dictates the grounding-line ice flux and grounding line stability, whereas the inferred viscosity maps inform the prediction of rifts. Both are essential for predicting the future mass loss of the Antarctic Ice Sheet.