Why Assumptions & Uncertainties in Bed DEMs Matter: A Geothermal Example from Haynes Glacier, West Antarctica

Subglacial digital elevation models (DEMs) are a critical boundary condition in glaciological modeling for hypothesis testing. Interpreting the results of this glaciological modeling requires careful consideration of statistical methods, data integrity, and error quantification when generating the bed DEMs used as a constraining boundary condition.  Here we use various DEMs to model ice flow over a suite of hypothesized geothermal anomalies, at Thwaites/Haynes Glaciers, West Antarctica, described in Bott et al. 2023 & 2024, to assess the impact of DEM selection on hypothesis testing for these potential melt anomalies.

Commonly used DEMs (e.g. Bedmap 2 and BedMachine), while robust on a continental level, do not in general represent the finer-scale variation in bed topography, which has a significant impact on ice dynamics from the ice divide all the way to the grounding line. 

Bedmap 2 uses a thin plate spline algorithm with iterative finite difference interpolation, producing a smooth interpolation of subglacial topography. This method has merit in providing a general topographic shape; however, the smaller-scale basal roughness is virtually erased, and notable inaccuracies exist in areas of high topographic relief, such as the edges of deep troughs and subglacial valleys. 

BedMachine, constrained by its mass-balance ice flow model must, by nature, assume where basal ice is sliding vs. where ice is frozen to the bed. This creates inherent circular logic in modeling, whereby one cannot constrain basal melt and basal sliding using BedMachine’s topography - since the thermodynamic state of basal ice has already been assumed by the DEM’s mass-balance model. 

To combat these issues, novel geostatistical and machine-learning methods for generating high-fidelity DEMs, which incorporate simulations of smaller-scale basal roughness, have been developed by Goff et al. 2014, Graham et al. 2017, and Dr. Emma MacKie at the University of Florida’s Gator Glaciology Lab. But all methods, regardless of their sophistication, come with assumptions, biases, and uncertainties. And with Bedmap 3 and BedMachine 4 not far from becoming publicly available, the selection of DEMs has never been wider. Here we use the potential for basal melting beneath Haynes/Thwaites Glaciers as a framework for understanding how underlying assumptions, methodologies, and uncertainties associated with the spectrum of bounding DEMs impact our confidence in the hypothesis tests based on this ice sheet modeling. 

Here we focus on the Thwaites/Haynes Glacier System of West Antarctica, which has two properties that make it ideal for assessing the impact of bed DEMs on ice sheet modeling results. 1) The presence of a trough-dominated basal morphology, characterized by heterogeneous geothermal flux provides ample conditions for testing the spatial distribution of basal sliding vs. plastic flow. And 2) a richness of volcaniclastic internal reflection horizons to be used as indicators of mass losses from basal melting through time.