Skip to main content
eScholarship
Open Access Publications from the University of California

Understanding Regional Ice Sheet Mass Balance: Remote Sensing, Regional Climate Models, and Deep Learning

  • Author(s): Mohajerani, Yara
  • Advisor(s): Velicogna, Isabella
  • et al.
Abstract

The Antarctic and Greenland ice sheets are experiencing significant mass change with heterogeneous spatial and temporal characteristics and global consequences such as sea level rise affecting millions of people in low-lying coastal areas. Advances in large-scale satellite remote-sensing, modeling, and machine learning have ushered a new era of improved monitoring and understanding of these changes. In this dissertation, we analyze the mass balance of glaciers across the ice sheets at basin and sub-basin scales using satellite gravimetric data from the Gravity Recovery and Climate Experiment (GRACE) mission using a novel regionally-optimized mascon methodology, as well as Mass Budget Method (MBM) estimates from grounding line discharge measurements and surface mass balance from regional climate models. We find that Totten and Moscow University glaciers in the marine sector of East Antarctica, with a total 5-meter sea level rise potential, have been losing mass at a rate of 18.5±6.6 Gt/yr from April 2002 to August 2016. The MBM estimate obtained with RACMO2.3p1 (Regional Atmospheric Climate Model version 2.3 part 1) is in excellent agreement with GRACE at a sub-basin scale, while those obtained with RACMO2.3p2 and MAR (Modèle Atmosphérique Régional) version 3.6.41 show less negative trends. These results are robust with respect to Glacial Isostatic Adjustment (GIA) uncertainty. By extending this methodology to the Amery Ice Shelf drainage basin in East Antarctica, we find this basin is in balance and is also in agreement with MBM/RACMO2.3p1 at a sub-basin scale, while MBM/RACMO2.3p2 and MBM/MAR3.6.41 produce more positive trends. The discrepancies shown by RACMO2.3p2 and MAR3.6.41 in these regions of East Antarctica are attributed to larger mean monthly SMB magnitudes. By adjusting all models to have the same mean magnitude as RACMO2.3p1, all MBM time-series fall into agreement with the independent gravimetric data. Furthermore, we implement the regional optimization approach in the Getz Ice Shelf drainage basin in West Antarctica, where previous studies have shown disagreements between GRACE and MBM estimates, and find that by minimizing leakage in the GRACE estimate, all MBM estimates are in excellent agreement with the gravimetric result. The Getz Ice Shelf basin is found to have a mass loss rate of 22.9±10.9 Gt/yr with an acceleration of 1.6±0.9 Gt/yr2 from April 2002 to November 2015 (the common time-period with the MBM estimates). We use an ensemble of 128,000 GIA forward models to ensure the results are robust with respect to GIA uncertainty. Lastly, we focus on improving the monitoring and understanding of glacier dynamics by implementing a deep Convolutional Neural Network (CNN) to automatically delineate glacier calving fronts from Landsat imagery on the Greenland Ice Sheet. By training the network on Jakobshavn, Sverdrup, and Kangerlussuaq glaciers and testing it on Helheim glacier, we demonstrate that the performance of the network is comparable to that of a human investigator, with a mean CNN error of 1.97 pixels (96.3 meters) compared to a mean human error of 1.89 pixels (92.5 meters) on the same resolution images. Thus, we show that CNNs enable large-scale monitoring of glacier dynamics across the globe, which offers new possibilities for an improved understanding of the processes affecting the mass balance of glaciers. Ultimately, a better understanding of the ice sheets is crucial for a better assessment of the effects of a changing cryosphere and sea level rise around the globe.

Main Content
Current View