Ice Mass Loss Monitoring in the Canadian Arctic: A Study on the Filtering Methods with Release-05 GRACE Data

English Abstract:

The biggest challenge in GRACE gravity data analysis is the signal separation of the GRACE-observed integrated geophysical signals. Due to the geometry of the twin-satellite orbit noise is produced, which is described as long, linear features generally oriented from north to south in the GRACE maps of mass variability. These ‘stripes’ imply a high degree of correlation in the gravity field coefficients, so spatial noise filtering is needed. As the halfwidth of the filter is increased, the level of smoothing is increased and the amplitude of the stripes decreases, until the pattern of the geophysical signals become apparent, but the magnitude of the geophysical signals is reduced.

In this study, in order to remove the ‘striping’ effect, a series of miscellaneous filtering methods is applied, tested and evaluated by ice mass loss estimates from previous studies in the Canadian Arctic. The data used include 139 months of GFZ and CSR Release-05 (RL05) GRACE time-variable gravity coefficients, covering the time period 01/2003-10/2014. Monthly gravity coefficients are converted to mass changes and corrected for the Glacial Isostatic Adjustment (GIA) effect.

The filtering methods that have been tested are: (1) Wiener filter (the signal and the error covariance depend only on the spherical harmonic degree), (2) post-processing decorrelation filter followed by Gaussian smoothing (convolution with an isotropic Gaussian smoothing kernel), and (3) non-isotropic smoothing (decorrelation) using the k-filtered five DDK filters RL05 GRACE solutions. We conclude that the general performance of isotropic filters is weak for ice applications.