Granular decoherence precedes failure of glacial ice mélange

Predicting impending failure in disordered systems is a principal goal in many fields ranging from earthquake detection to glassy, granular, and mechanical metamaterials. In most cases, particle and bond-level information plays a crucial role in predicting failure, yet this level of detail is often unavailable for complex geophysical systems. In flowing granular materials, machine learning techniques and acoustic emissions analyses demonstrate precursors to failure; yet, real-time detection remains an elusive goal. Here we show that failure of ice mélange, a large-scale granular material that is pushed through fjords by tidewater glaciers, is preceded by a loss of coherent flow. By analyzing terrestrial radar data sampled every 3-minutes, we find that the spatial pattern of strain rates within ice mélange develops large-scale fluctuations as early as 1 hour before an iceberg calving event. We also use a particle dynamics model to show how these fluctuations are likely due to buckling and rearrangements of the quasi-two-dimensional ice mélange. Our results directly implicate ice mélange as a mechanical inhibitor of calving along tidewater glaciers, and further demonstrate the potential for real-time detection of failure in geophysical granular materials.