An improved parameter filtering approach for processing GRACE gravity field models using first-order Gauss–Markov process

Removing stripe noise from the GRACE (Gravity Recovery and Climate Experiment) monthly gravity field model is crucial for accurately interpreting temporal gravity variations. The conventional parameter filtering (CPF) approach expresses the signal components with a harmonic model while neglecting non-periodic and interannual signals. To address this issue, we improve the CPF approach by incorporating those ignored signals using a first-order Gauss–Markov process. The improved parameter filtering (IPF) approach is used to filter the monthly spherical harmonic coefficients (SHCs) of the Tongji-Grace2018 model from April 2002 to December 2016. Compared to the CPF approach, the IPF approach exhibits stronger signals in low-degree SHCs (i.e., degrees below 20) and lower noise in high-order SHCs (i.e., orders above 40), alongside higher signal-to-noise ratios and better agreement with CSR mascon product and NOAH model in global and basin analysis. Across the 22 largest basins worldwide, the average Nash–Sutcliffe coefficients of latitude-weighted terrestrial water storage anomalies filtered by the IPF approach relative to those derived from CSR mascon product and NOAH model are 0.90 and 0.21, significantly higher than 0.17 and − 0.71, filtered by the CPF approach. Simulation experiments further demonstrate that the IPF approach yields the filtered results closest to the actual signals, reducing root-mean-square errors by 30.1%, 25.9%, 45.3%, 30.9%, 46.6%, 32.7%, 39.6%, and 38.2% over land, and 2.8%, 54.4%, 70.1%, 15.3%, 69.2%, 46.5%, 40.4%, and 23.6% over the ocean, compared to CPF, DDK3, least square, RMS, Gaussian 300, Fan 300, Gaussian 300 with P4M6, and Fan 300 with P4M6 filtering approaches, respectively