[Paper Review] Physics-Informed Machine Learning On Polar Ice: A Survey

Zesheng Liu, Younghyun Koo+1 more|arXiv (Cornell University)|Apr 30, 2024

Arctic and Antarctic ice dynamics5 citations

TL;DR

A comprehensive survey of physics-informed machine learning (PIML) for polar ice, detailing methods to combine physical ice models with data-driven learning and outlining challenges and opportunities.

ABSTRACT

The mass loss of the polar ice sheets contributes considerably to ongoing sea-level rise and changing ocean circulation, leading to coastal flooding and risking the homes and livelihoods of tens of millions of people globally. To address the complex problem of ice behavior, physical models and data-driven models have been proposed in the literature. Although traditional physical models can guarantee physically meaningful results, they have limitations in producing high-resolution results. On the other hand, data-driven approaches require large amounts of high-quality and labeled data, which is rarely available in the polar regions. Hence, as a promising framework that leverages the advantages of physical models and data-driven methods, physics-informed machine learning (PIML) has been widely studied in recent years. In this paper, we review the existing algorithms of PIML, provide our own taxonomy based on the methods of combining physics and data-driven approaches, and analyze the advantages of PIML in the aspects of accuracy and efficiency. Further, our survey discusses some current challenges and highlights future opportunities, including PIML on sea ice studies, PIML with different combination methods and backbone networks, and neural operator methods.

Motivation & Objective

• Motivate the study of polar ice dynamics and its impact on global climate and sea level rise.
• Review traditional physical and data-driven models used for land ice and sea ice.
• Propose a taxonomy for physics-informed machine learning methods in polar ice.
• Analyze advantages, limitations, and future opportunities of PIML in polar ice applications.

Proposed method

• Classify PIML approaches into three integration methods: physics-informed loss functions, physics-aware model architectures, and physics-informed training strategies.
• Survey physical laws governing land ice and sea ice (mass and momentum conservation, EVP, and sea ice dynamics) and their computational representations.
• Review data-driven models (MLP, CNN, RNN, GNN, GAN) applied to land ice and sea ice topics.
• Compare physics-based, data-driven, and PIML methods, highlighting accuracy and efficiency implications.
• Discuss case studies and methodological exemplars illustrating PIML usage in polar regions.

Experimental results

Research questions

• RQ1 What are the existing physics-informed machine learning approaches applied to polar ice?
• RQ2 How do physics-informed methods compare to traditional physical and data-driven models in accuracy and efficiency for polar ice problems?
• RQ3 What are the main challenges and future opportunities for PIML in land ice and sea ice contexts?
• RQ4 What taxonomy best organizes polar-ice PIML methods by how physics is incorporated?
• RQ5 What are the emergent directions (e.g., neural operators, backbones, combination strategies) in this field?

Key findings

• PIML leverages physical laws as constraints to ensure physically meaningful predictions while benefiting from data-driven learning.
• Three representative integration methods are loss-function tuning, architecture adjustment, and training-strategy design.
• Sea ice and land ice modeling involve complex physics (Stokes flows, BP, SSA, SIA, EVP formulations) that impact model choice and computational cost.
• Data-driven models (MLP, CNN, RNN, GNN, GAN) have been applied across ice thickness, concentration, front detection, and ice flow emulation, with varying data requirements.
• The survey highlights challenges such as data scarcity in polar regions and computational costs, and discusses opportunities like neural operators and multi-fidelity data fusion.