Data filling of terrestrial water storage anomaly during the gap period of GRACE/GRACE-FO: a case study of global typical basins

The long-term monitoring of TWSA (Terrestrial Water Storage Anomaly) is of great scientific significance to the research on water circulation and the allocation of water resources. Therefore, the problem caused by the data window period between GRACE (Gravity Recovery and Climate Experiment) and GRACE-FO (GRACE Follow-On) must be solved. In this paper, SSA (Singular Spectrum Analysis) +ARMA (Autoregressive Moving Average) is used to realize an iterative prediction and gap compensation of TWSA for the data window period. The TWSA of GRACE SH (Spherical Harmonic) inversion between January 2003 and December 2012 are as training samples and the TWSA between January 2013 and August 2016 as true values. Under this premise, through three approaches : ARMA, Wavelet Neural Network (WNN), SSA and SSA+ARMA, forecast tests are respectively carried out in eight typical basins: AZRB (Amazon River Basin), DNRB (Danube River Basin), GNRB (Ganges River Basin), MSRB (Mississippi River Basin), NGRB (Niger River Basin), VGRB (Volga River Basin), YNSRB (Yenisei River Basin) and YZRB (Yangtze River Basin). The accuracy of each approach is measured by RMSE (Root Mean Square Error), NSE (Nash-Sutcliffe Efficiency Coefficient) and R (Correlation Coefficient). The results show that prediction appears best in NGRB because there are strong periodic signals in TWSA time series. Also among the three approaches, SSA is shown to have the highest precision with the RMSE values of most basins less than 4 cm, the relevant NSE over 0. 75 and the R above 0. 85. So it is demonstrated that SSA+ARMA is excellent in identifying and extracting effective information from complicated signals. Following the first test, the TWSA from January 2003 to August 2016 are taken as training samples, and SSA approach is used to predict the TWSA from September 2016 to February 2020 and fill the data gaps. Through a validation comparing data from the TWSA of GRACE SH inversion, Mascons and GLDAS (Global Land Data Assimilation System), the filling results of TWSA have a high accuracy. Moreover, the results turn out more consistent with the TWSA of GRACE-FO SH inversion, and the RMSE values of most basins are less than 4 cm, the relevant NSE values are above 0.8 and the R values are above 0.9. To some extent, the consistency with Mascons and GLDAS data also proves the validity of the results. In conclusion, the research demonstrates that it is feasible and effective to use SSA + ARMA prediction to realize a data filling of TWSA during the gap period of GRACE/GRACE-FO.