Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving Greenland Bed Topography Mapping with Uncertainty-Aware Graph Learning on Sparse Radar Data

Bayu Adhi Tama, Homayra Alam, Mostafa Cham, Omar Faruque, Jianwu Wang, Vandana Janeja

TL;DR

This work tackles the challenge of reconstructing Greenland's subglacial bed topography from sparse radar observations by introducing GraphTopoNet, a graph-based framework that fuses surface covariates with gradient and polynomial trend augmentations. It employs a three-layer GCN to propagate information across a spatial grid and uses a uncertainty-aware hybrid loss that combines radar-confidence weighting, BedMachine reference guidance, and epistemic regularization via Monte Carlo dropout, with dynamic balancing of loss terms. Across three Greenland sub-regions, GraphTopoNet demonstrates superior accuracy, high structural fidelity (SSIM/PSNR), and near-unity R^2, including robust spatial extrapolation to unseen areas. The approach yields actionable, uncertainty-quantified bed maps suitable for sea-level projections and climate-risk assessment, and generalizes to other sparse geophysical domains with heterogeneous supervision.

Abstract

Accurate maps of Greenland's subglacial bed are essential for sea-level projections, but radar observations are sparse and uneven. We introduce GraphTopoNet, a graph-learning framework that fuses heterogeneous supervision and explicitly models uncertainty via Monte Carlo dropout. Spatial graphs built from surface observables (elevation, velocity, mass balance) are augmented with gradient features and polynomial trends to capture both local variability and broad structure. To handle data gaps, we employ a hybrid loss that combines confidence-weighted radar supervision with dynamically balanced regularization. Applied to three Greenland subregions, GraphTopoNet outperforms interpolation, convolutional, and graph-based baselines, reducing error by up to 60 percent while preserving fine-scale glacial features. The resulting bed maps improve reliability for operational modeling, supporting agencies engaged in climate forecasting and policy. More broadly, GraphTopoNet shows how graph machine learning can convert sparse, uncertain geophysical observations into actionable knowledge at continental scale.

Improving Greenland Bed Topography Mapping with Uncertainty-Aware Graph Learning on Sparse Radar Data

TL;DR

This work tackles the challenge of reconstructing Greenland's subglacial bed topography from sparse radar observations by introducing GraphTopoNet, a graph-based framework that fuses surface covariates with gradient and polynomial trend augmentations. It employs a three-layer GCN to propagate information across a spatial grid and uses a uncertainty-aware hybrid loss that combines radar-confidence weighting, BedMachine reference guidance, and epistemic regularization via Monte Carlo dropout, with dynamic balancing of loss terms. Across three Greenland sub-regions, GraphTopoNet demonstrates superior accuracy, high structural fidelity (SSIM/PSNR), and near-unity R^2, including robust spatial extrapolation to unseen areas. The approach yields actionable, uncertainty-quantified bed maps suitable for sea-level projections and climate-risk assessment, and generalizes to other sparse geophysical domains with heterogeneous supervision.

Abstract

Accurate maps of Greenland's subglacial bed are essential for sea-level projections, but radar observations are sparse and uneven. We introduce GraphTopoNet, a graph-learning framework that fuses heterogeneous supervision and explicitly models uncertainty via Monte Carlo dropout. Spatial graphs built from surface observables (elevation, velocity, mass balance) are augmented with gradient features and polynomial trends to capture both local variability and broad structure. To handle data gaps, we employ a hybrid loss that combines confidence-weighted radar supervision with dynamically balanced regularization. Applied to three Greenland subregions, GraphTopoNet outperforms interpolation, convolutional, and graph-based baselines, reducing error by up to 60 percent while preserving fine-scale glacial features. The resulting bed maps improve reliability for operational modeling, supporting agencies engaged in climate forecasting and policy. More broadly, GraphTopoNet shows how graph machine learning can convert sparse, uncertain geophysical observations into actionable knowledge at continental scale.

Paper Structure

This paper contains 22 sections, 6 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Proposed Framework
  5. Prior Physical Model (BedMachine)
  6. Feature Augmentation
  7. Graph Representation of Spatial Grid
  8. Graph Learning Architecture
  9. Hybrid Loss Function
  10. Experiments
  11. Datasets
  12. Experimental Settings
  13. Evaluation Metrics
  14. Baselines
  15. Results
  16. ...and 7 more sections

Figures (2)

  • Figure 1: Overview of the proposed GraphTopoNet framework. Surface-derived features (e.g., elevation, velocity, SMB) are processed and encoded over a spatial graph. Gradient and trend surface augmentations enhance the input representation. The graph is passed through a multi-layer Graph Convolutional Network (GCN) to predict bed topography. Supervision is provided by both radar observations and the BedMachine map, with uncertainty-aware loss balancing to guide learning in sparse or noisy regions.
  • Figure 2: Spatial visualization of BedMachine-derived bed topography with overlaid radar observation points for each sub-region, such as Upernavik Isstrøm (a), Hayes (b), and Kangerlussuaq (c). The figure illustrates the spatial coverage of radar data in relation to the underlying bed elevation.