[Paper Review] Domain-Decomposed Lagrangian Data Assimilation for Drifting Sea-Ice Floe Dynamics

Danyang Li, John Taylor+1 more|arXiv (Cornell University)|Feb 20, 2026

Arctic and Antarctic ice dynamics0 citations

TL;DR

The paper presents a domain-decomposed ensemble transform Kalman filter (ETKF) data assimilation framework for Lagrangian sea-ice floe dynamics, achieving better global ocean flow reconstruction than a full-domain baseline by local ETKF updates in subdomains with Gaussian blending.

ABSTRACT

Sea ice dynamics are crucial to the global climate system, yet traditional continuum (e.g., viscous-plastic) models often fail to represent the discrete floe interactions that dominate in the marginal ice zone. Lagrangian discrete element methods (DEMs) resolve floe-scale physics more realistically, but their high particle counts make ensemble data assimilation (DA) more expensive. We consider a highly-simplified floe model and propose a scalable, domain-decomposed DA framework that couples Lagrangian particle observations with an ensemble transform Kalman filter (ETKF) to recover the underlying ocean flow field in a multiscale setting. The Eulerian domain is first partitioned into subdomains. We then impose an ETKF in each subdomain to recover the local fine-scale ocean features. A Gaussian-weighted blending step then reconstructs a globally consistent flow field across subdomain boundaries. Numerical experiments demonstrate consistently better skill scores that are characterised by normalised root mean square error (NRMSE) and pattern correlation coefficients (PCC), compared to the global and expensive DA baseline. Results suggest that the domain-decomposed DA method is an alternative, scalable approach for particle-based sea-ice floe dynamics and ocean flow recovery.

Motivation & Objective

• Motivate the need for floe-scale data assimilation to capture discrete floe interactions in marginal ice zones.
• Propose a scalable domain-decomposed DA framework that couples Lagrangian floe observations with ETKF.
• Recover a globally consistent ocean flow field from local subdomain updates via Gaussian blending.
• Demonstrate the method's accuracy and scalability over a full-domain baseline in a simplified floe model.

Proposed method

• Partition the Eulerian domain into non-overlapping subdomains and run ETKF analyses locally.
• Select a reduced local floe observation set in each subdomain by prioritising floes near the subdomain centre.
• Update local floe states and local Fourier coefficients of the ocean model within each subdomain.
• Map updated Fourier coefficients to physical space to obtain local velocity fields and apply Gaussian blending across subdomains.
• Fuse subdomain velocity fields with Gaussian weights to reconstruct a globally coherent ocean flow field.
• Use an idealised Lagrangian DEM for floe dynamics with one-way coupling to an ocean Fourier-mode model.

Experimental results

Research questions

• RQ1 Can a domain-decomposed ETKF approach provide accurate reconstructions of the ocean flow field from sparse Lagrangian floe observations?
• RQ2 How does subdomain partitioning (e.g., 2x2 vs 4x4) affect accuracy, boundary artefacts, and computational cost?
• RQ3 Does Gaussian blending effectively preserve large-scale structures while dampening subdomain-edge noise?
• RQ4 How does the domain-decomposed method compare to a full-domain data assimilation baseline in terms of NRMSE, PCC, and runtime?

Key findings

• The domain-decomposed method consistently improves skill scores (NRMSE and PCC) over the full-domain baseline across observation budgets.
• 2x2 subdomain decomposition often yields better accuracy than 4x4 for comparable observation budgets due to better cross-subdomain coherence.
• Gaussian blending helps maintain global flow coherence and reduces boundary artefacts while enabling parallel subdomain updates.
• Increasing the number of observed floes improves accuracy, with diminishing returns at higher budgets.
• In some configurations, decomposed DA achieves similar or better accuracy at lower computational cost than the full-domain approach.
• Visual reconstructions show closer alignment to ground truth and reduced high-error regions for the decomposed method.