Inferring subglacial topography using physics informed machine learning constrained by two conservation laws

Krishna, Mansa; Cheng, Gong; Morlighem, Mathieu

doi:10.5194/tc-20-3533-2026

Articles | Volume 20, issue 6

https://doi.org/10.5194/tc-20-3533-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/tc-20-3533-2026

© Author(s) 2026. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 20, issue 6

Research article

|

19 Jun 2026

Research article |

| 19 Jun 2026

Inferring subglacial topography using physics informed machine learning constrained by two conservation laws

Mansa Krishna, Gong Cheng, and Mathieu Morlighem

Data sets

CReSIS Radar Depth Sounder Data CReSIS http://data.cresis.ku.edu/

IceBridge BedMachine Greenland, Version 6 M. Morlighem et al. https://doi.org/10.5067/6B6B225B8V2D

Model code and software

mansakrishna23/BedMappingPINN: BedMappingPINN (v3.0.0) M. Krishna https://doi.org/10.5281/zenodo.20182294

ISSMteam/PINNICLE: v0.1 G. Cheng et al. https://doi.org/10.5281/zenodo.14889235

Ice-sheet and Sea-level System Model source code, v4.23 r27696 ISSM Team https://doi.org/10.5281/zenodo.7850841

Short summary

Estimates of the Greenland Ice Sheet’s contribution to sea level rise are affected by uncertainties in the bed topography. Traditional, physics-based methods for inferring the bed elevation are limited to fast-flowing areas of the ice sheet. We use machine learning models informed with two physical laws to infer the bed elevation for different regions in Greenland, showing that this method can be used to infer the bed elevation in slower-moving, sparsely surveyed regions of the ice sheet.