Improved multichannel singular spectrum analysis for post-processing GRACE monthly gravity field models

Abstract

SUMMARY Multichannel singular spectrum analysis (MSSA) is a powerful tool to extract spatiotemporal signals and filter errors from the noisy time-series of monthly gravity field models from the satellite data of gravity recovery and climate experiment (GRACE). Since the GRACE monthly gravity models are missed about 17 months, we develop an improved MSSA approach, which can directly process the incomplete time-series without either data interpolation or iteration. The time-series of 14-yr (2002.04–2016.08) monthly gravity field models of CSR-RL06 up to degree and order 60 are analysed with improved MSSA compared to the MSSA with linear data interpolation and iteration MSSA. By using our improved MSSA approach, the first 11 principal components derived can capture 91.18 per cent of the total variance, higher than 85.80 and 86.44 per cent of the total variance, derived by linear interpolation MSSA and iteration MSSA, respectively. The ratios of the latitude weighted RMS over the land and ocean signals are used to evaluate the efficiency of eliminating noise by the MSSA approach. For improved MSSA, the mean RMS ratio of land and ocean signals of all available months is higher than linear interpolation and iteration MSSA, which indicates that improved MSSA can suppress noise more efficiently and extract more geophysical signals from real GRACE data. Furthermore, the 50 repeated experiments show that all the root mean squared errors and mean absolute errors derived by our improved MSSA are smaller than other MSSA approaches. Moreover, the improved MSSA performs still better than other MSSA based approaches for the cases of large data gaps.