Investigating differences between GRACE mass change estimation methods using their inherent sensitivity kernels

Estimates of mass changes from GRACE spherical harmonic (SH) solutions arrive at different results even for the same regions and timeframes. These differences can be attributed to two parts of the estimation. The first part is the preparation of a SH input dataset based on the GRACE solutions. This includes amendments to low-degree components and corrections for modelled geophysical signals. The second part, and topic of this presentation, is the estimation method itself. Studies assessing these estimation methods were mostly based on the resulting mass change estimates from a limited number of test signals. We propose to use the inherent sensitivity kernels (SKs) of estimation methods for comparison and assessment of methods. The SK describes the weighting function that is used to integrate a surface mass density representation of the prepared input dataset. It can be represented in either the spatial or SH domain. Alternative terms for the SK used in previous studies include 'averaging kernel', 'averaging function' or 'weight function'. For methods of the direct approach the SK is obvious. These methods directly construct the SK, mostly starting from modification of a simple region function. For methods of the inverse approach the SKs are less obvious. In this approach patterns of mass changes are defined which are then fitted to observations of the gravity field variations. However, SKs are also inherent to inverse methods and may be made explicit. In fact, certain implementations of both approaches have identical SKs when rigorously incorporating the same signal and error covariance information. Under this conditions both approaches are equivalent. Based on the SKs it is straight-forward to assess leakage errors and GRACE error propagation not only in the resulting mass change estimation. Publishing the SKs associated to a method also enables any user to investigate the method without deeper knowledge of the implementation details. We illustrate this by presenting SKs for our implementation of four different methods to estimate Greenland Ice Sheet mass changes. The SKs show similarities as well as striking differences which can be attributed to the underlying differences in the methods.