Methodology:

We used static grids of ice surface elevation from the REMA mosaic (Howat and others, 2019) and ice thickness (from which bed elevation is inferred; Morlighem and others, 2020; 2022) to predict the subglacial water flow routing pathways based on hydropotential gradients. Hydropotential gradients control where water is driven to flow by gravitational potential (bed elevation) and ice overburden pressure (Shreve, 1972), given by equation 1:

del-phi = rho . g . del-z . del-Pw

where phi is the hydropotential gradient, rho is the density of water, g is gravitational acceleration, z is bed elevation, and Pw is water pressure. Pw assumes effective pressure (N) is zero, which is a common assumption in subglacial hydrology where effective pressure is challenging to measure. Instead Pw is calculated solely from the depth of the ice column above. Flow routing pathways were modelled based on the hydropotential surface using the Python package, pysheds release 0.3.5 (Bartos, 2020). We conditioned the hydropotential surface by pit and depression filling and resolved flats. Flow direction and accumulation used the default directional mapping that specifies the routing algorithm. We extracted flow paths with flow accumulations >1,000 upstream cells. Flow paths were clipped at the grounding line (Depoorter and others, 2013) and filtered to only include flow paths intersecting the region of interest (Whillans Ice Plain, West Antarctica, the contributory Whillans Ice Stream and neighbouring Kamb and Mercer Ice Streams). Computation was conducted using a ~32 GB server on CryoCloud cloud-computing hub (Snow and others, 2023).

Temporal coverage:

* We used static grids of ice surface elevation from BedMachine Antarctica, Version 3 (Morlighem and others, 2020; 2022) that uses the REMA mosaic (Howat and others, 2019), which was created with sub-meter resolution Maxar optical satellite imagery acquired across multiple austral summer seasons between 2009 and 2021.

* We used static grids of ice thickness from BedMachine Antarctica, Version 3 (Morlighem and others, 2020; 2022). BedMachine Antarctica uses a variety of input datasets with temporal coverage spanning 1961 to 2020.

Location:

The dataset covers the region of interest, the Whillans Ice Plain, plus the upstream ice drainage basin of Whillans Ice Stream and neighbouring Kamb and Mercer Ice Streams as defined in the refined basins in the MEaSUREs Antarctic Boundaries for IPY 2007-2009 from Satellite Radar, Version 2 dataset (Mouginot and others, 2017).

CRS: Polar stereographic projection (?EPSG:3031?)

Data collection:

To generate the predicted subglacial flow paths, we used the Python package, pysheds release 0.3.5 (Bartos, 2020).

Data quality:

Ice-surface elevations from REMA (Howat and others, 2019) have relatively low error, typically less than 1 m. Sources of error in BedMachine Antarctica's estimates of ice thickness and thus bed elevation include error in ice velocity direction and magnitude, error in surface mass balance, and ice thinning rates. Below grounded ice, the error in mass conservation derived ice thickness and bed elevation ranges from 36 m to exceeding 200 m based on the density radar sounding coverage (Morlighem and others, 2020; 2022). Within the region of subglacial flow path mapping, BedMachine Antarctica has a mean error of 74.8 m and a maximum error of 786.0 m, which is most prominent in the upper catchment of Kamb Ice Stream.