Efficient basin scale filtering of GRACE satellite products

Abstract

International audienceThe Gravity Recovery And Climate Experiment (GRACE) satellite mission provides time-variable gravity fields that are commonly used to study regional and global terrestrial total water storage (TWS) changes. These estimates are superimposed by different error sources such as the north–south stripes in the spatial domain and spectral/spatial leakage errors, which should be reduced before use in hydrological applications. Although different filtering methods have been developed to mitigate these errors, their performances are known to vary between regions. In this study, a Kernel Fourier Integration (KeFIn) filter is proposed, which can significantly decrease leakage errors over (small) river basins through a two-step post-processing algorithm. The first step mitigates the measurement noise and the aliasing of unmodelled high-frequency mass variations, and the second step contains an efficient kernel to decrease the leakage errors. To evaluate its performance, the KeFIn filter is compared with commonly used filters based on (i) basin/gridded scaling factors and (ii) ordinary basin averaging kernels. Two test scenarios are considered that include synthetic data with properties similar to GRACE TWS estimates within 43 globally distributed river basins of various sizes and application of the filters on real GRACE data. The KeFIn filter is assessed against water flux observations through the water balance equations as well as in-situ measurements. Results of both tests indicate a remarkable improvement after applying the KeFIn filter with leakage errors reduced in 34 out of the 43 assessed river basins and an average improvement of about 23.38% in leakage error reduction compared to other filters applied in this study

Understanding the global hydrological droughts of 2003–2016 and their relationships with teleconnections

“…The smoothed fields are then converted to TWS changes following Wahr et al (1998). Basin average values for the 156 river basins of Figure 1 and their errors are estimated following (e.g., Khaki et al, 2018a). Our computations cover the complete mission period of 2003-2016, where Figure 2 shows the standard deviations of basin averaged GRACE TWS, their errors, and the signal to noise ratio within each basin of Figure 1.…”

“…This is also applied to other data sets, thus, all available data records have been synchronized. Besides, since other data sets have a spatial resolution different than that of GRACE L2 data, they are converted to the spectral domain and truncated at spherical harmonic degree and order 90 and basin averages are computed following Khaki et al (2018a). Figure 1.…”

Forecast of seasonal water availability in Central Asia with near-real time GRACE water storage anomalies

“…Dedicated post-processing of GRACE data for small basins of interest as proposed, e.g.,. by Longuevergne et al (2010) or Khaki et al (2018), may reduce the magnitude of filter and leakage errors in the final storage time series for these areas, but such an approach will be of limited use for the present forecasting study due to the gaps in the time series, thus suffering the same constraints in linear modelling as the standard monthly solutions. Expectations, albeit of limited scope, of a slightly higher spatial resolution of GRACE-FollowOn data based on its laser ranging interferometer (Flechtner et al 2016) may open pathways for transferring the proposed seasonal discharge forecasting method with GRACE predictors to smaller catchments.…”

Nonparametric Data Assimilation Scheme for Land Hydrological Applications

“…Three degree 1 coefficients (C10, C11, and S11) and degree 2 and order 0 (C20) coefficient are replaced by those of Swenson et al () and that of Cheng and Tapley (), respectively. Further, we apply the DDK2 smoothing filter (Kusche et al, ) to mitigate a colored/correlated noise in the coefficients (see also Khaki, Forootan, Kuhn, Awange, Longuevergne, & Wada, ), and thereafter convert them into 1° × 1° TWS fields following Wahr et al (). The mean TWS for the study period is taken from the W3RA model and is aggregated to the GRACE TWS change time series to reach absolute values related to W3RA (Zaitchik et al, ).…”