An Adaptive Regularized Filtering Approach for Processing GRACE Time-Variable Gravity Field Models

Objectives The gravity recovery and climate experiment time-variable gravity field model suffers from significant north-south striping errors due to insufficient east-west sampling, background force model errors, and inadequate solving strategies. These errors severely limit its application. While the official decorrelation and denoising kernel （DDK) regularization filtering algorithm is widely used, it has notable shortcomings:(1) The regularization parameters are empirically determined and do not account for monthly variability, using the same parameters for each month.(2) The Tikhonov regularization method over-regularizes low-frequency components and under-regularizes high-frequency components. To address these issues, we propose an adaptive regularization filtering method.

Methods This method estimates low-frequency components using least squares (without regularization), applies Tikhonov regularization to mid-frequency components, and truncates high-frequency components. Each frequency band and its regularization parameters are optimized based on the minimum mean square error criterion and processed separately each month.

Results The proposed method is applied to the ITSG-Grace2018 time-variable gravity field data with a maximum degree of 96, spanning from April 2002 to June 2017. Experimental results show that our proposed method achieves higher spatial resolution of mass anomalies and aligns more closely with the three official mascon products compared to the classic DDK filtering.

Conclusions Simulation experiments further validate that the mass anomalies obtained by this method are closer to the simulated true values.