Error characterization of GNSS vertical land displacements for use in GNSS-GRACE joint inversion

Quantification of uncertainty in surface mass change signals derived from GNSS measurements poses challenges, especially when dealing with large datasets with continental or global coverage. Our aim is to assign weights to GNSS estimates of vertical land displacement (VLD), which will be used in a future joint solution with GRACE observations. Thus, we study the structure and quantify the uncertainty present in VLD estimates derived from 3045 GNSS stations distributed across the continental US. Monthly means of daily positions are available for 15 years. First, we remove outliers by performing a 3σ test, and data screening via a number of correlation metrics between the input GNSS VLD estimates and external validation datasets (i.e., VLD predicted from GRACE/GRACE-FO and hydrology models). Afterwards, we employ various processing schemes to characterize the uncertainty of VLD through stochastic modeling and quantification of the spatially correlated errors. In particular, we test for white, colored and spatially correlated noise. When only white noise is considered nearly 30% of the stations exhibit noise level < 2 mm, 65% noise between 2-4 mm and 5% noise > 4 mm. In case of colored noise, error smaller than 2 mm, between 2-4 mm and >4mm, is mapped in 20%, 60% and 20% of the stations, respectively. Spatially correlated noise is in family but slightly smaller in magnitude compared to colored noise.