Icebergs, jigsaw puzzles and genealogy: Automated multi-generational iceberg tracking and lineage reconstruction

Abstract

Tabular icebergs calve from ice shelves and glaciers in Antarctica, Greenland, and northern Ellesmere Island. These ‘ice islands’, as they are referred to in the Arctic, drift, melt, and fragment, contributing fresh water and nutrients to the ocean, influencing circulation, carbon cycling and biodiversity in ways that remain poorly understood. Icebergs also pose risks to shipping, and maritime infrastructure. Improved understanding of iceberg drift and fragmentation will reduce uncertainties in climate simulations and operational hazards. This study presents the first comprehensively validated, scalable iceberg tracking approach and the first that is capable of reconstructing iceberg ‘lineages’ (here used to describe life histories including sources, where that source is a larger iceberg) through fragmentation events. This method enables a comprehensive reconstruction of iceberg paths from calving to their eventual disintegration, allowing for monitoring and source attribution across their life cycle. We propose CryoTrack, an unsupervised approach based on iceberg geometry that is agnostic to data source or delineation method. The system requires only vector outlines. Initially, icebergs are linked across timesteps when their shapes remain similar, forming ‘tracklets’. When significant shape changes occur, fragmented ‘child’ icebergs are linked to their ‘parents’ using a fuzzy geometric assembly method based on dynamic time warping, akin to assembling a jigsaw puzzle without image data. This approach reconstructs full iceberg lineages back to their calving origin. We evaluate system performance using manually tracked iceberg outlines in the Canadian Arctic. Standard tracking metrics and custom iceberg specific metrics assess its accuracy in scientific and operational contexts. Our approach achieves excellent tracking of icebergs with an overall tracking accuracy of 0.98 and 94 % of iceberg area are correctly linked to source when icebergs are last observed. This system contributes to the need for scalable iceberg monitoring. It enhances understanding of iceberg behaviours, impacts, and fragmentation, supporting process based and data driven predictive modelling for environmental and operational applications.