Multi-scale total water storage variations from fusion of GRACE and GLDAS Noah data using wavelet analysis

The observations of total water storage variations (TWSV) from GRACE satellites have a coarse spatial resolution of ~300 km. Hydrological and land surface models, providing TWSV with higher spatial resolutions like 50 km, offer an opportunity to improve the coarse spatial resolution of GRACE data. We present a fusion approach based on wavelet multiresolution analysis to combine TWSV data from GRACE and GLDAS Noah model. We first decompose the TWSV maps from GRACE and GLDAS into their building blocks at various spatial scales, examine their signal characteristics, and then combine complementary spatial features at the wavelet coefficient level. The spectral nature of our approach enables us to easily combine the large-scale components from GRACE with small-scales information from GLDAS Noah and produce a unified, multi-scale TWVS dataset which has the advantages of both input datasets. We show that the spatial signal frequency spectrum of the fusion dataset matches that of GRACE data at low spatial frequencies, but that of GLDAS data at high spatial frequencies, indicating that the algorithm provides a shrinkage free combination of both input datasets. For a case study in contiguous United States, we inspect fused TWSV dataset in the spatial domain. Despite having detailed features originating from GLDAS, the fused dataset accurately quantifies the water budget and its long-term trend, when averaged over large basins during 2003–2015 similar to GRACE. We use the high-resolution surface soil moisture from SMOS satellite and snow water equivalent from SNODAS to demonstrate the improvement in representation of small-scale features associated with soil moisture and snow (i.e., the two water storage components that GLDAS Noah TWS is comprised of) in our fused dataset. In both cases, our fused TWSV dataset shows notable higher correlations with these independent data compared to GLDAS.