MKMansa Krishna

Mansa Krishna

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Mansa Krishna has a total of 40 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

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

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, surface mass balance, and ice thinning rates in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    Quantifying the uncertainty in Greenland’s contribution to sea level rise due to the bed topography

    by Mansa Krishna, Mathieu Morlighem, Danielle Mangini, Youngmin Choi

    Uncertainty in the future contribution of the ice sheets to sea level rise associated with different climate forcings has been well studied in the most recent Ice Sheet Model Intercomparison Project for CMIP6 (ISMIP6). Similarly, the uncertainty due to differences in basal sliding laws or calving laws has also been thoroughly investigated. However, the uncertainty due to the bed topography has not yet been rigorously quantified. The majority of the models within the ISMIP6 ensemble use a single bed topography map, BedMachine Greenland, thereby hindering our ability to better understand how uncertainties in the bed topography affect the overall uncertainty in future projections of sea level rise. To address this, we follow the methodology from Castleman et al. (2022) and create an ensemble of 32 bed topography maps with realistic bed roughness by perturbing the BedMachine bed topography using discrete wavelet decomposition techniques. We update the initial bed topography in ice sheet models from Choi et al., (2021), which provide Greenland-wide, high-resolution, data constrained projections that include calving dynamics, and run projections out to 2300. Though we expect northwest and central west Greenland glaciers to contribute more to sea level rise than other glaciers, we find that models initialized with BedMachine bed topographies tend to overestimate mass loss in these regions. We also find that the addition of bed roughness reduces the future contribution of the ice sheet to sea level rise over the 21st century, but to a lesser extent than the deep, wide subglacial basins of Antarctica. Lastly, we also determine that the uncertainty in the future contribution of the Greenland Ice Sheet due to different climate forcings and the uncertainty due to the bed topography are comparable at the end of this century, however the uncertainty due to climate forcings dominates in the long run.

  • Open Access

    Reply on RC2

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, and the apparent mass balance in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    Reply on RC1

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, and the apparent mass balance in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    Comment on egusphere-2025-3964

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, and the apparent mass balance in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    Comment on egusphere-2025-3964

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, and the apparent mass balance in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    DeepTopoNet: A Framework for Subglacial Topography Estimation on the Greenland Ice Sheets

    by Bayu Adhi Tama, Mansa Krishna, Homayra Alam, Mostafa Cham, Omar Faruque, Gong Cheng, Jianwu Wang, Mathieu Morlighem, Vandana Janeja


  • Open Access

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

    by Mansa Krishna, Gong Cheng, Mathieu Morlighem

    Subglacial topography beneath the Greenland Ice Sheet is a fundamental control on its dynamics and response to changes in the climate system. Yet, it remains challenging to measure directly, and existing representations of the subglacial topography rely on a limited number of observations. Although the use of mass conservation and the development of BedMachine Greenland substantially improved the representation of the bed topography, this approach is limited to fast-flowing sectors and is less effective in regions with complex, alpine topography. As an alternative to traditional numerical methods, recent work has explored using Physics Informed Neural Networks (PINNs), constrained by only one physical law, to solve forward and inverse problems in ice sheet modeling. Building on this work, we assess three PINN frameworks constrained by distinct conservation laws, showing that PINNs informed with a single conservation law are not sufficient for regions with sparse measurements and complex topographies. To that end, we introduce a novel approach that involves coupling two conservation laws within a PINN framework to infer the subglacial topography and test this approach for three regions with distinct environments in Greenland. This PINN is trained with both the conservation of mass and an approximation of the conservation of momentum (the Shelfy-Stream Approximation), which allows us to simultaneously infer the ice thickness and basal shear stress using observations of ice velocities, surface elevation, and the apparent mass balance in a mixed inversion problem. We compare the predicted ice thickness to ground-truth ice-penetrating radar measurements of ice thickness, showing that the PINN informed with two conservation laws is capable of inferring ice thickness in sparsely surveyed regions. Furthermore, comparisons of predicted bed topographies with BedMachine Greenland show that this approach is capable of discovering new bed features in slower-moving regions and in regions of complex topography, highlighting its potential for better constraining the bed topography of the Greenland Ice Sheet.

  • Open Access

    A Python library for solving ice sheet modeling problems using physics-informed neural networks, PINNICLE v1.0

    by Gong Cheng, Mansa Krishna, Mathieu Morlighem

    Predicting the future contributions of the ice sheets to sea-level rise remains a significant challenge due to our limited understanding of key physical processes (e.g., basal friction, ice rheology) and the lack of observations of critical model inputs (e.g., bed topography). Traditional numerical models typically rely on data assimilation methods to estimate these variables by solving inverse problems based on conservation laws of mass, momentum, and energy. However, these methods are not versatile and require extensive code development to incorporate new physics. Moreover, their dependence on data alignment within computational grids hampers their adaptability, especially in the context of sparse data availability in space and time. To address these limitations, we developed PINNICLE (Physics-Informed Neural Networks for Ice and CLimatE), an open-source Python library dedicated to ice sheet modeling. PINNICLE seamlessly integrates observational data and physical laws, facilitating the solution of both forward and inverse problems within a single framework. PINNICLE currently supports a variety of conservation laws, including the Shelfy-Stream Approximation (SSA), MOno-Layer Higher-Order (MOLHO) models, and mass conservation equations, for both time-independent and time-dependent simulations. The library is user-friendly, requiring only the setting of a few hyperparameters for standard modeling tasks, while advanced users can define custom models within the framework. Additionally, PINNICLE is based on the DeepXDE library, which supports widely used machine learning packages such as TensorFlow, PyTorch, and JAX, enabling users to select the backend that best fits their hardware. We describe here the implementation of PINNICLE and showcase this library with examples across the Greenland and Antarctic ice sheets for a range of forward and inverse problems.


  • Open Access

    From Imagery to Insight: Machine Learning for Grain-Scale Sediment Analysis on Mars

    by Shreya Gandhi, Ayden Hayes, Mansa Krishna, Emma Rogers, Ari Koeppel

    Introduction: Rovers have traversed over 100 kilometers of the Martian surface, assembling an expansive archive of over one million high-quality imagery of sedimentary features in diverse geologic settings.1 These images record fine sediment grains (0.5-2 mm in diameter) shaped by their billions of years of surface processes. Studying grain morphology offers insight into sediments’ origins and deposition, crucial to reconstruct regional climatic histories.2 Grain size and form indicate transport history and paleo-depositional environments, making them fundamental expressions of Mars’ climatic and hydrological evolution.3,4 While previous studies classify individual rock outcrops along the rovers’ traverses, broader questions about sedimentary grain variability across the Martian surface remain unanswered. This project addresses that gap by applying deep learning to quantify grain morphometry across rover imagery, enabling us to extrapolate a novel large-scale sedimentary analysis. Data & Methods: We assess imagery from NASA’s Spirit, Opportunity, Curiosity, and Perseverance rovers which are each equipped with optical cameras meant for geologic investigation. The rovers have collectively returned across more than 13,000 sols that capture a wide range of morphological features including volcanic, aeolian, lacustrine, deltaic, and fluvial systems.3,4 We classified grains using five widely-accepted metrics in Martian sedimentology: grain size distribution (D50, D84), axial ratio (Ra), circularity (IR), convexity (C), and roundness (P).4,5,6,7 These characteristics allow us to infer transport histories, and determine environments—alluvial fan/fluvial, deltaic, lacustrine, aeolian, and crater floor/wall deposits. To automate grain form identification, we use a deep learning image segmentation model and high-quality unobstructed images from NASA’s Planetary Data Sciences Imaging node.8 Our work builds on the precedent of other domain-adapted deep learning models on Mars9 but shifts the focus to finer geologic units. We adapted the Segment Anything Model (SAM2), a vision transformer by Meta AI Research that uses global self-attention to contextualize local features and perform general-purpose segmentation.10 Unlike conventional convolutional neural networks, SAM2 is adept in complex, cluttered scenes, ideal for Martian surface imagery, and pixel-level segmentation is possible without retraining the model extensively. This circumvents the need for densely labeled training data, the likes of which do not exist for Martian sediments, and helps enable flexible, semi-supervised or even unsupervised workflows that can still produce geologically meaningful outputs. This significantly lowers the barrier to entry for planetary-scale sediment analysis and allows us to scale insights without exhaustive manual input. We further adapted SAM2 using a transfer learning strategy, drawing upon Segment Every Grain, an open-source Earth-based framework that applies SAM2 to cobble-sized sediment segmentation.11 We retrain the model on a subset of annotated rover imagery, enabling us to specialize the model according to the Martian surface. Reliably detected grains’ morphometric features can then be recorded, and linked to metadata about rover location and image characteristics to populate datasets for spatial and temporal analyses. Results & Discussion: By automating grain-scale segmentation, we anticipate a large-volume dataset that allows us to compare sediment characteristics across diverse Martian terrains and enables large-scale sediment provenance and transport studies that are otherwise infeasible via manual analysis. Our preliminary results suggest that SAM2 is effective in the identification of grains: the figure below demonstrates an image of Martian sediment and the identification of grains within the image that are then classified according to morphometric qualities. Figure 1: Left demonstrates the image before segmentation, showing a dense cluster of pebble-sized clasts. The right panel demonstrates several correctly-identified grains from which morphometric quantities were extracted using SAM2. Anticipated results include a novel geospatial distribution of styles of grain processing and their correlations to inferred depositional systems. This opens new frontiers for regional climatic reconstructions in a new light. Beyond Mars, this work contributes broadly to planetary science, sedimentology, and machine vision. Such automated grain-scale analysis could be extended to other planetary bodies. Likewise, the methodology could help advance Earth-based geoscience, including fluvial transport modeling, coastal erosion studies, and landslide prediction.3 All code, models, and processed datasets from this project will be made open-source to support future research and maximize scientific return from planetary imagery, encouraging community-driven development. References: [1] Rice, M. S., et al. (2017), Geologic overview of the Mars Science Laboratory rover mission at the Kimberley, Gale crater, Mars, J. Geophys. Res. Planets, 122, 2–20. [2] Wadell, H. (1932). Volume, Shape, and Roundness of Rock Particles. The Journal of Geology, 40(5), 443–451. [3] Attal, M., and J. Lavé (2009), Pebble abrasion during fluvial transport: Experimental results and implications for the evolution of the sediment load along rivers, J. Geophys. Res., 114, F04023. [4] Szabó, T., Domokos, G., Grotzinger, J., et al (2015), Reconstructing the transport history of pebbles on Mars. Nat Commun 6, 8366. [5] Yingst, R. A., et al. (2013), Characteristics of pebble- and cobble-sized clasts along the Curiosity rover traverse from Bradbury Landing to Rocknest, J. Geophys. Res. Planets, 118, 2361–2380. [6] Palucis, M. C., W. E. Dietrich, A. G. Hayes, R. M. E. Williams, S. Gupta, N. Mangold, H. Newsom, C. Hardgrove, F. Calef III, and D. Y. Sumner (2014), The origin and...