Comparison of GRACE(-FO) data with geophysical models at a small spatial scale

One of the applications of satellite gravimetry data from GRACE and GRACE Follow-On (GFO) missions is a comparison with various geophysical models that provide information about mass re-distribution in the Earth system. An example is regional climate models describing the Surface Mass Balance (SMB) of ice sheets and glaciers. Possible goals of such a comparison are model validation, as well as recovery of signals not captured by models. This comparison may be a challenging task because it frequently requires that GRACE-/GFO- based estimates are produced at a small spatial scale and with a high accuracy. These requirements conflict with intrinsic limitations of GRACE/GFO data. Even if random noise is suppressed using dedicated algorithms, the accuracy of the obtained GRACE/GFO-based estimates is inevitably reduced due to signal leakage.

We propose a novel scheme for a comparison of GRACE/GFO-based Spherical Harmonic Coefficients (SHCs) with geophysical models. To mitigate the “internal” signal leakage, we adopt a fully consistent data inversion. This includes, among others, a transformation of geophysical estimates into Spherical Harmonic Coefficients (SHCs). Then, all the sets of SHCs are inverted into the spatial domain using the same data weighting. To mitigate the signal leakage from outside, we estimate mass anomalies over a global set of patches (mascons). To reduce non-uniqueness of the inversion (which may manifest itself, e.g., as Gibbs phenomenon), we exploit available prior knowledge, such as the fact that mass anomalies typically show only a minor variability over the ocean and the accumulation zone of the Greenland Ice Sheet (GrIS). This prior knowledge is introduced in the form of a first-order Tikhonov regularization. To find the optimal regularization parameters, we train the inversion scheme using realistically simulated synthetic data. To capture the geometry of coastlines accurately, we use patches of a relatively small size (about 40x40 km at the latitude of the Greenland’s central part). Since this results in a huge number of unknown parameters, we invert SHCs iteratively, using the Preconditioned Conjugate gradient method.

We apply the proposed scheme to compare seasonal mass anomalies based on GRACE/GFO data and on SMB estimates from the Regional Atmospheric Climate Model RACMO2.3p2. The comparison is limited to the Greenland’s coastal zone, which includes the ablation zone, as well as tundra and isolated ice caps outside it. We split the coastal zone into 12 regions. The linear size of each region is only a few hundred km, i.e., close to the theoretical limit of resolution achievable from GRACE/GFO data. We identify regions where mass anomaly time-series of the two types show a good agreement (< 2 cm RMS in terms of equivalent water heights) and regions where the discrepancies are larger. In most cases, those discrepancies are likely caused by an under- or over-estimation of meltwater runoff, as well as the buffered water storage en route to the ocean, which is sensed by GRACE/GFO, but is not included into SMB models.