Correlated error reduction in GRACE data over Greenland using extended empirical orthogonal functions

Abstract

Surface mass change estimates from Gravity Recovery and Climate Experiment (GRACE) spherical harmonic solutions are contaminated by north‐south stripe noise due largely to aliasing of high‐frequency variations into monthly samples. These meridional stripes are especially troubling for ice mass balance studies of the Greenland Ice Sheet (GrIS) where large ice mass variations are known to occur along north‐south trending coastlines. By assuming that mass variations and noise have different patterns in both space and time over Greenland, we use extended empirical orthogonal functions (EEOFs) to filter out this noise. The method is compared with a conventional approach, by examining both continent‐wide estimates and regional changes. GRACE results are compared with independent regional estimates derived from a climate model. The EEOF filter is effective at separating ice mass change signals from meridional stripe noise, with better rejection of high temporal frequency noise and less signal attenuation and spatial smoothing compared to a conventional method. We use EEOF‐filtered GRACE data to examine regional seasonal variations. Consistent with surface and other data, results show ice mass loss along the west, southwest, and east coasts during summer and gain in these regions during winter. In addition, there is summer ice mass gain in the central region of the GrIS.

Global sea level change signatures observed by GRACE satellite gravimetry

“…Mascon solutions, for example, have provided spatially gridded data with significantly reduced signal leakages by effectively separating signal between land and oceans 9 – 11 . However, a recent study showed that mascon data have unrealistic signals present along the coastline of Greenland 12 , indicating that this approach to leakage correction is imperfect. An effective way to correct this leakage problem is an algorithm called forward modelling 13 .…”

Sparse DDK: A Data-Driven Decorrelation Filter for GRACE Level-2 Products

“…They are different from empirical decorrelation filters [8][9][10], which also find wide applications. There are also many other filters, such as [11][12][13][14][15][16][17][18], but these are out of the scope of this study since we mainly focus on DDK-like filters in this work. Usually, two types of input are necessary to design a DDK filter, namely the noise covariances and the signal covariances.…”

High‐Resolution GRACE Monthly Spherical Harmonic Solutions

“…To suppress the correlated noise, one option is to use de‐striping filtering (Swenson & Wahr, 2006). In addition, other filters such as Gaussian smoothing (Jekeli, 1981; Wahr et al., 1998), DDK filter (Kusche et al., 2009), Wiener filter (Sasgen et al., 2006) and Empirical orthogonal function (EOF) filter (Eom et al., 2017; Wouters & Schrama, 2007) are also applicable to mitigate the impacts of the correlated noise. Although these filtering methods can significantly reduce the correlated noise and users can freely choose a suitable one according to the demand of different geophysical applications, many studies showed that the spatial resolution of filtered solutions was generally reduced (Watkins et al., 2015).…”

“…As a monitor of time‐variable mass redistribution of the Earth, the Gravity Recovery and Climate Experiment (GRACE) mission was launched in March 2002 (Tapley et al., 2004) and stopped service in October 2017 due to battery issue. The GRACE mission provided us an unprecedented understanding of the Earth's surface mass transports at global and regional scales, for example, ocean mass transports (Kusche et al., 2016), ice mass loss over Antarctica and Greenland (Eom et al., 2017; Schrama et al., 2014; Velicogna & Wahr, 2013), water storage variations (Abd‐Elbaky & Jin, 2019; Forootan et al., 2017; Long et al., 2017; Rodell et al., 2009), and Earthquake deformation (Han et al., 2006; Wang et al., 2012). The quantification of the Earth's mass transports is practically conducted based on GRACE‐derived gravity field models.…”