Subglacial Conditions Estimated From Unsupervised Clustering Analysis of Radar Bed-Echo Shape

Abstract

Methods that can uniquely constrain subglacial properties are essential for understanding the role of basal processes in modern glacier evolution. Here, we demonstrate a new framework to characterize ice-sheet basal conditions using a basin-scale ($>$10,000 line-km) airborne radio-echo sounding (RES) survey collected in the 2013–14 austral summer over Whillans Ice Plain, Crary Ice Rise, and Ross Ice Shelf. We train a batched, unsupervised K-Means clustering algorithm to group radar bed echoes within the study region, targeting similarities in the shape (as opposed to amplitude) of 1-D scattering profiles at the bed (here, referred to as bed-echo waveforms). We additionally explore the effects of the transmitted beam and survey design on cluster outcomes. We interpret the meaning behind clustered bed-echo waveforms and their spatial distributions in the study region to assess the capabilities of our framework for characterizing the subglacial environment over a basin-scale RES survey. Our findings indicate that clustering the basal ice portion of the bed-echo waveforms yields distinctly shaped clusters and glaciologically meaningful spatial distributions. Estimates of basal environment characteristics from the clustered RES survey used in this study reveal regional contrasts in basal properties across the ice plain and indications of localized processes that influence ice dynamics (e.g., material entrainment). Our bed-echo clustering workflow offers a radar attenuation-independent approach for identifying basal processes under ice sheets––where uncertainty in path effects makes interpreting scattering phenomena challenging––and holds broad potential for application in multi-mission, repeat-track, orbital, and multi-decadal RES surveys.