Mansa Krishna

Short Biography

 

---

Research Summary

Mansa’s research aims to improve estimates of subglacial bed topography beneath Greenland’s glaciers, a key factor controlling ice flow and sea level rise projections. Because direct measurements are sparse, she uses Physics-Informed Neural Networks (PINNs) that incorporate both mass and momentum conservation laws to infer ice thickness more accurately. This method solves a 2D inverse problem by using surface velocity, elevation, slope, and limited thickness data to estimate both ice thickness and basal slipperiness. Unlike previous approaches, her model performs well even in slower-moving glacier regions, where traditional methods tend to be less accurate. By applying this framework to glaciers like Upernavik and Narssap, Mansa demonstrates improved agreement with radar-based observations, advancing the reliability of ice sheet models.

---

Publications

• G. Cheng, M. Krishna, and M. Morlighem. 2025. A Python library for solving ice sheet modeling problems using Physics Informed Neural Networks, PINNICLE v1.0. EGUsphere 2025, (2025), 1–26. https://doi.org/10.5194/egusphere-2025-1188 [preprint]

---

Proposal Defense

Successfully Defended On: February 19, 2025

Title: New Bed Features Beneath the Greenland Ice Sheet and their Implications for Ice Sheet Model Projections

Abstract

Subglacial bed topography is a fundamental control on ice dynamics and its response to climate change. Yet, it remains difficult to measure and existing descriptions of the bed topography rely on a limited number of direct observations. Poor constraints on the bed topography are a key cause for uncertainty in model projections. While the use of mass conservation and the development of BedMachine Greenland dramatically improved the representation of the subglacial bed topography, this approach is limited to fast-moving regions. For the first chapter of my thesis, I propose inferring the bed topography using Physics-Informed Neural Networks (PINNs) constrained with two conservation laws: mass conservation and momentum conservation. By coupling mass conservation with momentum conservation, I aim to show that PINNs are capable of inferring the bed topography in slower moving regions of the Greenland Ice Sheet (GrIS), where traditional inverse methods are less effective. For the second chapter of my thesis, I will be using this approach to infer the bed topography for the entire GrIS while also generating a map of the uncertainty in the bed topography. For the third and final chapter of my thesis, I will be assessing how uncertainties in the bed topography affect the spread of ice sheet model projections. Ultimately, I aim to show that improving the accuracy of the bed topography of the GrIS will improve the predictive skills of model simulations and reduce the spread of uncertainty in sea level rise projections.

Committee

• Dr. Mathieu Morlighem – Advisor and Committee Chair (Dartmouth College)
• Dr. C. Brenhin Keller – Committee Member (Dartmouth College)
• Dr. Marisa Palucis – Committee Member (Dartmouth College)
• Dr. Ching-Yao Lai – Committee Member (Stanford University)

Last updated 16 May 2025