Abstract
Monitoring the flow velocity of Greenland’s outlet glaciers is essential for quantifying ice sheet mass loss and its response to global climate change. Synthetic Aperture Radar (SAR) offset-tracking is a widely used technique for this purpose; however, rapid changes in surface scattering properties—often caused by summer melting or snowfall—frequently lead to decorrelation and subsequent data gaps. This study proposes a robust data processing workflow using Sentinel-1 SAR imagery to generate high-resolution velocity time-series. By integrating redundant image pairs (6, 12, and 18-day intervals) and employing an iterative Singular Value Decomposition (SVD) solver, the method effectively bridges temporal gaps and suppresses noise. Applied to the Petermann Glacier from 2018 to 2020, the workflow produced a continuous 6-day interval velocity series that outperforms existing products in terms of noise reduction, spatiotemporal continuity, and coverage during peak melt seasons.
Introduction
The dynamics of Greenland's outlet glaciers are a critical indicator of the ice sheet's health. Accurate monitoring of ice flow velocity allows researchers to calculate mass balance and predict future sea-level rise. SAR offset-tracking has become the standard method for measuring these movements by tracking the displacement of surface features between satellite passes.
Despite its utility, offset-tracking faces significant challenges. The primary issue is decorrelation: when the glacier surface undergoes rapid physical changes—such as intense summer melting, heavy snowfall, or calving events—the radar signature changes so drastically that matching features between images becomes impossible. This results in velocity maps characterized by significant noise and "holes" where data is missing, precisely during the periods of highest glaciological interest.
Methodology
To overcome the limitations of standard offset-tracking, this research introduces a five-step iterative workflow designed to clean raw data and synthesize a continuous time-series.
The workflow consists of:
- Initial Denoising: Raw tracking results undergo morphological "opening" to remove speckle noise, connectivity analysis to eliminate isolated pixel clusters, and adaptive median filtering.
- Baseline Correction: Low-resolution reference products (e.g., GIMP) are used to identify stable areas for calibration. Flow direction constraints are applied to filter out physically unrealistic "backward" movements.
- Redundant Pairing: Instead of relying solely on consecutive 6-day image pairs, the method incorporates 12 and 18-day intervals. This redundancy ensures that if a short-term pair fails due to surface melting, longer-interval pairs can still capture the underlying movement.
- Iterative SVD: A system of equations is constructed where the 6-day velocities are the unknowns and the redundant pairs are observations. This system is solved using Singular Value Decomposition (SVD). The process is iterative: residuals are calculated to identify and discard outliers, and the system is re-solved until convergence.
- Spatiotemporal Filtering: A final 3D median filter is applied to ensure the resulting velocity field is smooth across both spatial and temporal dimensions.
Data and Study Area
The study focuses on the Petermann Glacier, a major outlet glacier in North Greenland. The primary dataset includes 178 scenes of Sentinel-1 A/B Interferometric Wide (IW) mode Single Look Complex (SLC) imagery spanning 2018 to 2020. Reference data for calibration and benchmarking include the Greenland Ice Mapping Project (GIMP) products, as well as existing velocity products from CPOM (6/12-day single-track) and PROMICE (12/24-day multi-track).
Results and Discussion
The proposed method successfully generated a continuous 6-day interval velocity time-series for the Petermann Glacier.
Key findings include:
- Improved Continuity: Compared to the CPOM product, the new workflow produces velocity fields that are significantly smoother and exhibit higher effective data coverage. The use of redundant pairs effectively "glues" together periods that would otherwise be lost to decorrelation.
- High Resolution and Accuracy: The results match the accuracy of the multi-track PROMICE product while offering double the temporal resolution (6 days versus 12/24 days).
- Seasonal Insights: The method is particularly robust during summer and autumn, periods when standard tracking often fails. The data revealed distinct seasonal cycles, showing that the Petermann Glacier flowed faster and for longer durations in the summers of 2019 and 2020 compared to 2018.
Conclusion
The integration of redundant image pairing and iterative SVD provides a powerful framework for monitoring glacier dynamics. While some data gaps persist during extreme melt events, the method significantly enhances our ability to observe high-frequency changes in ice flow. Future work will explore the integration of multi-track SAR data and optical imagery (e.g., Landsat or Sentinel-2) to further bolster reliability during challenging seasons and scale the process for larger ice sheet regions.
Summary updated