Global and Planetary Change 109 (2013) 30–37

Contents lists available at ScienceDirect

Global and Planetary Change

journal homepage: www.elsevier.com/locate/gloplacha

Characterization of spatio-temporal patterns for various GRACE- and

GLDAS-born estimates for changes of global terrestrial water storage
Tao Yang a,b,⁎, Chao Wang a, Zhongbo Yu a,c, Feng Xu d
a
State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Center for Global Change and Water Cycle, Hohai University, Nanjing 210098, China
b
State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi, China
c
Department of Geoscience, University of Nevada Las Vegas, Las Vegas, NV 89154-4010, USA
d
College of Computer and Information, Hohai University, Nanjing 210098, China

a r t i c l e i n f o a b s t r a c t

Article history: Since the launch in March 2002, the Gravity Recovery and Climate Experiment (GRACE) satellite mission has
Received 10 October 2012 provided us with a new method to estimate terrestrial water storage (TWS) variations by measuring earth
Received in revised form 23 June 2013 gravity change with unprecedented accuracy. Thus far, a number of standardized GRACE-born TWS products
Accepted 10 July 2013
are published by different international research teams. However, no characterization of spatio-temporal
Available online 18 July 2013
patterns for different GRACE hydrology products from the global perspective could be found. It is still a big
Keywords:
challenge for the science community to identify the reliable global measurement of TWS anomalies due to our
GRACE-born hydrology products limited knowledge on the true value. Hence, it is urgently necessary to evaluate the uncertainty for various global
GLDAS estimates of the GRACE-born TWS changes by a number of international research organizations. Toward this end,
Intercomparison this article presents an in-depth analysis for various GRACE-born and GLDAS-based estimates for changes of
Change of terrestrial water storage global terrestrial water storage. The work characterizes the inter-annual and intra-annual variability, probability
Global scale density variations, and spatial patterns among different GRACE-born TWS estimates over six major continents,
and compares them with results from GLDAS simulations. The underlying causes of inconsistency between
GRACE- and GLDAS-born TWS estimates are thoroughly analyzed with an aim to improve our current knowledge
in monitoring global TWS change. With a comprehensive consideration of the advantages and disadvantages among
GRACE- and GLDAS-born TWS anomalies, a summary is thereafter recommended as a rapid reference for scientists,
end-users, and policy-makers in the practices of global TWS change research. To our best knowledge, this work is the
first attempt to characterize difference and uncertainty among various GRACE-born terrestrial water storage
changes over the major continents estimated by a number of international research organizations. The results
can provide beneficial reference to usage of different GRACE hydrology products to study TWS changes in different
regions of the world.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction for TWS measurements with unprecedented accuracy (Tapley et al.,

2004). Till far, GRACE-born products have been increasingly used for a
Terrestrial water storage (TWS) plays an essential role within wide variety of purposes on hydrology and water resources (Ramillien
the global water cycle and on climate (Famiglietti, 2004), and it is one et al., 2008; Schmidt et al., 2008). Some examples include evaluating
of the extremely important resources related to economic and societal hydrological fluxes such as evapotranspiration (Rodell et al., 2004a;
development. Monitoring and accurate estimate of TWS variations is Ramillien et al., 2006; Rodell et al., 2011), estimating rates of groundwater
fundamental to watch and predict droughts, floods and other natural depletion (Swenson et al., 2008b; Fukuda et al., 2009; Rodell et al., 2009)
hazards under the condition of global climate change. However, cost and improving parameter estimation in land surface models (Yirdaw
of intensive and expensive labor and materials have seriously hindered et al., 2009; Lo et al., 2010).
the development of networks to measure TWS changes, especially at TWS changes can be accurately derived from the terrestrial gravity
continental and global scales. field variations measured by the GRACE satellite mission. Nevertheless,
Satellite observations of time-variable gravity from the Gravity Recov- the retrieval is not straightforward, and different processing methods
ery and Climate Experiment (GRACE) mission provided new approaches will result in different estimates of the GRACE-born TWS products.
Although a considerable number of studies have been carried out widely
to investigate TWS changes over different parts of the globe using the
⁎ Corresponding author at: State Key Laboratory of Hydrology-Water Resources
and Hydraulic Engineering, Center for Global Change and Water Cycle, Hohai University,
GRACE-born data, only a few GRACE studies have been implemented at
Nanjing 210098, PR China. Tel.: +86 25 83786017. continental or global scales (Ramillien et al., 2004; Wahr et al., 2004;
E-mail address: [email protected] (T. Yang). Schmidt et al., 2006; Grippa et al., 2011). Moreover, as a key point, no

0921-8181/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.gloplacha.2013.07.005
 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37 31

Table 1
The GRACE-born estimates of global TWS change.

Product name Institute Countries involved Reference or/and sponsorship Spatial grid Spatial resolution Temporal resolution

JPL-GRACE Tellus NASA Jet Propulsion Laboratory USA Swenson and Wahr (2006) 1° × 1° 300 km 1 month
DEOSS DMT V 1b Delft University of Technology, Netherlands, China Klees et al. (2008) Liu et al. (2010) 0.5° × 0.5° 200 km 1 month
Wuhan University
CNES-GRGS V 2 Space Geodesy Research Group France Bruinsma et al. (2010) 1° × 1° 400 km 10 days
CU, Boulder The University of Colorado USA Swenson and Wahr (2002, 2006), 1° × 1° 300 km 1 month
Swenson et al. (2008a)

assessment on the reliability or uncertainty for various global estimates Wahr, 2002). The scaling coefficients provided for each 1° bin
of the GRACE-born TWS changes by a number of international research of the GRACE gridded data are used to compensate bias and
organizations has been found so far. This seriously limits further in- leakage and restore much of the energy removed by truncation,
depth utilization of the global GRACE-born products in hydrology. With destriping and Gaussian smoothing processes (Swenson and
this regard in mind, the study hereby strives to answer how reliable Wahr, 2006).
different estimates of GRACE-born TWS change are in different conti- (2) The DEOS Mass Transport Model (DMT) monthly solutions by
nents, through a comprehensive inter-comparison with a variety of the University of Delft (http://www.lr.tudelft.nl) are also based
estimates from the GRACE and Global Land Data Assimilation System on the decomposition into spherical harmonic coefficients but
(GLDAS) (Rodell et al., 2004b). This is because we are unable to identify to degree and order 120 (Liu et al., 2010). The data is available
the true value of global TWS currently. Meanwhile, the inter-annual and from February 2003 to August 2010. Computation of purely
intra-annual variability of different global GRACE-born and GLDAS-born dynamic orbits of GRACE satellites and monthly solutions with
TWS estimates are detected. The findings from this study can signifi- corresponding covariance matrices are executed iteratively, as
cantly contribute to identify reliable estimates of global GRACE-born described in detail by Liu (2008). The series of monthly solutions
TWS in different continents, and boost our current skills in monitoring is post-processed by applying statistically optimal Wiener filters
global TWS change under the climate change conditions using the based on full signal and noise covariance matrices. The signal
state-of-the-art GRACE technology. variances and solutions are computed iteratively, according to
the schemes addressed by Klees et al. (2008). Each monthly solu-
2. Data and method tion is further transformed into an equivalent water layer thick-
ness with units of meter as a function of latitude and longitude
2.1. Estimates of TWS changes by different GRACE-born products in spherical coordinates according to the method described by
Wahr et al. (1998).
Four different GRACE hydrology products (Table 1) employed in this (3) The GRGS-EIGEN-GL04 10 day models released by the Centre Na-
study are briefly described below. tional d'Études Spatiales (CNES) available from August 2002 to July
2011 are computed based on the GRACE GPS and K band range
(1) The monthly land water solutions from the GRACE Tellus website rate (KBRR) data and LAGEOS-1/2 SLR data (http://grgs.obs-mip.
(http://gracetellus.jpl.nasa.gov/data/gracemonthlymassgridsland/) fr/index.php/fre/Donnees-scientifiques/Champ-de-gravite/grace)
by the NASA Jet Propulsion Laboratory (JPL) with a spatial resolu- (Bruinsma et al., 2010). These gravity fields consist of a set of
tion of 1° × 1° are available from April 2002 to November 2011. normalized spherical harmonic coefficients from degree 2 up
These monthly GRACE gravity models are based on the U. Texas to degree and order 50 applying a stabilization method without
Center for Space Research RL version 4.0 gravity field coefficients additional filtering. Unlike land grid version “ss201008” monthly
(land grid version “ss201008”) truncated at degree 60 (Swenson data, GRGS products do not require additional filtering, hence
and Wahr, 2006). The degree 2 order 0 coefficients observed no scaling factor is applied for the GRGS data (Ramillien et al.,
by GRACE showed unreasonable variability, so were replaced 2008; Tregoning et al., 2008). The 10 day GRGS solutions were
with values derived from Satellite Laser Ranging (SLR) (Cheng converted to a monthly time series by taking the average values
and Tapley, 2004). The degree 1 coefficients are those derived in order to directly compare them with other GRACE solutions
by Swenson et al. (2008a) due to GRACE's inability to provide de- employed in our study.
gree 1 coefficients which represent geocenter motion (Chen (4) The error-corrected mass anomalies from the University of
et al., 2005a). The GRACE mass anomalies were further corrected Colorado (CU) real-time GRACE website (http://geoid.colorado.
for the postglacial rebound contributions generated by the high- edu/grace/grace.php) spans from June 2002 to November 2009.
latitude Pleistocene deglaciation according to the model of Gravity fields used at this site are truncated at degree 70. From
Paulson et al. (2007). A destriping filter was applied to the data each monthly solution, a time-mean field is removed to obtain
to reduce the anisotropic noise, which manifests itself as strong the TWS anomaly. The gravity field data are expressed in units
north–south stripes (Swenson and Wahr, 2006). After destriping, of millimeters of equivalent water thickness using relationship
the signal was further smoothed using a 300 km wide Gaussian between changes in the geopotential and changes in the associ-
filter to reduce high degree measurement errors (Swenson and ated surface-mass density. After the data are converted to mass

Table 2
Land Surface Models in GLDAS.

Model acronym Institute Countries involved Spatial grid Reference

CLM 2.0 The National Center for Atmospheric Research (NCAR) USA 1° × 1° Bonan et al. (2002), Dai et al. (2003)
MOSAIC v3 The National Aeronautics and Space Administration (NASA) USA 1° × 1° Koster and Suarez (1996)
NOAH 2.7 The National Oceanic and Atmospheric Administration (NOAA) USA 1° × 1° Chen et al. (1996), Koren et al. (1999), Ek et al. (2003)
VIC 4.0.4 University of Washington (UW) USA 1° × 1° Liang et al. (1994)
32Terrestrial Water Storage Anomaly (cm) T. Yang et al. / Global and Planetary Change 109 (2013) 30–37

Terrestrial Water Storage Anomaly (cm)

4 A) 4 B)
2
2

0
0

-2
-2

-4
-4
2

2
9

1
03

04

05

06

07

08

09

10

11

03

04

05

06

07

08

09

10

11
00

00

00

00

00

00

00

00
00

01

01

00

00

00

00

00

00

00

01

01
20

20

20

20

20

20

20

20

20

20

20

20

20

20

20

20

20

20
ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2
ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2
n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.

n.
Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju
Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju
Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja
Month Month

Terrestrial Water Storage Anomaly (cm)

20
Terrestrial Water Storage Anomaly (cm)

12 18
10
C) 16
14
D)
12
8 10
6 8
6
4 4
2
2 0
-2
0 -4
-6
-2 -8
-4 -10
-12
-6 -14
-16
-8 -18
-20
2

2
7

3
8

6
9

1
03

04

05

06

07

03
08

04

05
09

10

11

06

07

08

09

10

11
00

00

00

00

00

00
00

00
00

00

00

00
00

01

01

00

00

00

01

01
20

20

20

20

20

20
20

20

20
20

20

20

20

20

20

20

20

20
ly 2

ly 2

ly 2

ly 2

ly 2

ly 2
ly 2

ly 2
ly 2

ly 2

ly 2

ly 2
ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2

ly 2
n.

n.

n.

n.

n.

n.
n.

n.

n.
n.

n.

n.

n.

n.

n.

n.

n.

n.
Ju

Ju

Ju

Ju

Ju

Ju
Ju

Ju
Ju

Ju

Ju

Ju
Ju

Ju

Ju

Ju

Ju

Ju

Ju

Ju
Ja

Ja

Ja

Ja

Ja

Ja
Ja

Ja

Ja
Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja

Ja
Month Month
Terrestrial Water Storage Anomaly (cm)
Terrestrial Water Storage Anomaly (cm)

10 14

8
E) 12 F)
10
6 8
6
4
4
2 2
0 0
-2
-2 -4
-4 -6
-8
-6 -10
-8 -12
-14
-10
2

2
7

5
9

7
1

1
03

04

05

06

07

03
08

04

05
09

10

06

07
11

08

09

10

11
00

00

00

00

00

00
00

00

00

00

00
00

01

00

00
01

00

00

01

01
20

20

20

20

20

20
20

20

20
20

20

20

20
20

20

20

20

20
ly 2

ly 2

ly 2

ly 2

ly 2

ly 2
ly 2

ly 2

ly 2

ly 2

ly 2
ly 2

ly 2

ly 2

ly 2
ly 2

ly 2

ly 2

ly 2

ly 2
n.

n.

n.

n.

n.

n.
n.

n.

n.
n.

n.

n.

n.
n.

n.

n.

n.

n.
Ju

Ju

Ju

Ju

Ju

Ju
Ju

Ju

Ju

Ju

Ju
Ju

Ju

Ju

Ju
Ju

Ju

Ju

Ju

Ju
Ja

Ja

Ja

Ja

Ja

Ja
Ja

Ja

Ja
Ja

Ja

Ja

Ja
Ja

Ja

Ja

Ja

Ja
Month Month

JPL CNES DMT CU CLM Mosaic Noah VIC

Fig. 1. TWS anomalies in (A) Africa, (B) Asia, (C) Australia, (D) Europe, (E) North America, and (F) South America estimated from four GRACE products and four GLDAS model outputs.
The TWS anomalies represent averaged equivalent water thickness changes in the specific continent.

units, they are smoothed in the spectral domain using a Gaussian (3) The NOAH LSM (Chen et al., 1996; Koren et al., 1999; Ek et al.,
filter to reduce short-wavelength spectral noise (Wahr et al., 2003) is a basic component of the GLDAS. This LSM simulates
1998). soil moisture (both liquid and frozen), soil temperature, surface
temperature, snowpack depth, snowpack water equivalent (and
2.2. Estimates of TWS changes by different land surface models in GLDAS hence snowpack density), canopy water content, and the energy
flux and water flux terms of the surface energy balance and
Four estimates of TWS change by different land surface models surface water balance.
in GLDAS (Table 2) collected into this study are presented as following: (4) The Variable Infiltration Capacity model (VIC) (Liang et al., 1994)
is a macroscale hydrologic model that solves full water and energy
(1) The Community Land Model (CLM) (Bonan et al., 2002; Dai
et al., 2003) is the land model consisting of four components:
biogeophysics, hydrologic cycle, biogeochemistry and dynamic
Table 3
vegetation. The land surface is represented by 5 primary sub- Average RMSE of the residuals among different GRACE and GLDAS products and averaged
grid land cover types (glacier, lake, wetland, urban, vegetated) GLDAS results in the six major continents (in Units of water height, cm).
in each grid cell. The current version of the Community Land
CONTINENTS JPL CNES DMT CU CLM Mosaic Noah VIC
Model in GLDAS is CLM2.0.
(2) The Mosaic Land Surface Model (LSM) (Koster and Suarez, 1996) AFRICA 1.20 1.09 1.54 1.15 0.79 0.76 0.37 0.77
ASIA 1.56 1.70 1.51 1.65 0.55 0.80 0.63 0.66
constitutes a basic component of the GLDAS. The Mosaic LSM
AUSTRALIA 1.86 2.13 1.57 1.80 1.46 0.91 0.84 1.04
separates each grid cell into subgrid “tiles” based on vegetation EUROPE 3.22 3.43 5.14 3.87 2.69 2.49 1.22 1.73
class and then performs separate energy and water balance calcu- NORTH AMERICA 2.86 2.31 2.53 2.13 1.66 1.73 1.17 1.57
lations over each tile. The model includes three soil layers and SOUTH AMERICA 2.85 2.73 2.71 2.59 3.69 2.18 0.96 1.43
MEAN 2.26 2.23 2.50 2.20 1.81 1.48 0.87 1.20
a simple one-layer snow model in which melt rate is calculated
VARIANCE 0.57 0.55 1.63 0.75 1.18 0.48 0.09 0.16
as the residual of the surface energy balance.
 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37 33

6
4
A) Africa B) Asia
4

2
2
Range

Range
0
0

-2 -2

-4
JPL CNES DMT CU CLM Mosaic Noah VIC JPL CNES DMT CU CLM Mosaic Noah VIC

20

C) Australia 15
D) Europe
10

10

5
5
Range

Range
0
0
-5

-5 -10

JPL CNES DMT CU CLM Mosaic Noah VIC JPL CNES DMT CU CLM Mosaic Noah VIC

10

E) North America 15 F) South America

5 10

5
Range

Range

0
0

-5
-5

-10

-10
JPL CNES DMT CU CLM Mosaic Noah VIC JPL CNES DMT CU CLM Mosaic Noah VIC

JPL CNES DMT CU CLM Mosaic Noah VIC

Fig. 2. Box plots of TWS changes.

balances. VIC is a research model and in its various forms it (Europe Fig. 1D, North America Fig. 1E), October (South America,
has been applied to many watersheds and globally as well. Fig. 1F) and November (Australia, Fig. 1C). An abrupt high TWS is detected
The model now forms a basic part of the GLDAS. in Australia during 2010 to 2011 (Fig. 1C), when a series of floods hit
Australia, beginning in December 2010, primarily in the state of
Continental-scale averaged TWS changes are then computed from Queensland. However, GLDAS estimates indicate a more distinct seasonal
scaled GRACE global fields using cosine (latitude) weighting. TWS change cycle than GRACE observations, even in Africa (Fig. 1A) and Asia (Fig. 1B).
time series from different products are intercompared with GLDAS Moreover, a strong intra-annual variability can be found in GRACE
TWS to validate the reliability of different GRACE- and GLDAS-born TWS results, which could not be observed in GLDAS outputs. Grippa
products in estimating TWS changes in different continents. et al. (2011) have also reported these phenomena when they studied
land water storage variability over West Africa, but its cause remains
3. Results and discussion unclear.
The four different GRACE-born time series show good coherence with
3.1. Inter-annual and intra-annual variability for various GRACE- and each other. Table 3 provides the averaged root mean square (RMS) of the
GLDAS-born estimates residuals among different GRACE, GLDAS products and mean GLDAS
results in the six major continents. The agreement between different
3.1.1. Time series GRACE results is generally good considering their amplitude and phase.
Fig. 1 shows the resulting 10-yr time series of the monthly TWS South America shows the greatest seasonal variability (~14 cm), followed
anomaly as equivalent water heights (EWHs) estimated from four by Europe and North America.
GRACE hydrology products and four GLDAS models in the six continents As the amplitudes of the seasonal water storage variations is consid-
(excluding the Antarctic and Greenland). The time series have been ered, four GRACE products show some minor distinctions, with the DMT
recentered for each solution by removing the temporal mean over the often lower than the other estimates (e.g. Australia, Europe, South
common period 2003 to 2011. America). Temporal changes simulated from four GLDAS models
Except for Africa (Fig. 1A) and Asia (Fig. 1B), series (Fig. 1C–D) esti- are quite consistent, and the dry and wet phases are well represented.
mated from different GRACE products show an obvious seasonal cycle in From the time series we can identify that CLM outcomes show lower
TWS anomalies across the world. TWS peaks and troughs from GRACE amplitudes than other three models in Africa, Australia and South
and GLDAS generally appear at the same time, with maxima usually America, all in the southern hemisphere. Fig. 1 also suggests that
identified in March (Australia, Europe, Fig. 1C–D) and April (North GRACE observations are lower than GLADS estimates over Europe and
America, South America, Fig. 1E–F) and minima found in September North America.
34 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37

1.2 0.6
A) Africa 0.5
B) Asia
1.0

0.4
Probability Density

Probability Density
0.8

0.3
0.6

0.2
0.4
0.1
0.2
0.0
0.0
-0.1
-6 -4 -2 0 2 4 6 8 -4 -2 0 2 4
Terrestrial Water Storage Anomaly (cm) Terrestrial Water Storage Anomaly (cm)
0.6 0.6

0.5
C) Australia 0.5
D) Europe
0.4 0.4
Probability Density

Probability Density
0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0

-0.1 -0.1
-15 -10 -5 0 5 10 15 20 -30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30
Terrestrial Water Storage Anomaly (cm) Terrestrial Water Storage Anomaly (cm)
0.6 0.6

0.5
E) North America 0.5
F) South America
0.4 0.4
Probability Density

Probability Density

0.3 0.3

0.2 0.2

0.1 0.1

0.0 0.0

-0.1 -0.1
-15 -10 -5 0 5 10 15 20 -20 -10 0 10 20

Terrestrial Water Storage Anomaly (cm) Terrestrial Water Storage Anomaly (cm)

JPL CNES DMT CU CLM Mosaic Noah VIC

Fig. 3. Probability density estimates for TWS changes.

3.1.2. Box plots 3.1.3. Probability density plots

To quantify the difference of means and variances among various Fig. 3 demonstrates the probability density estimates of TWS change
GRACE- and GLDAS-born estimates of TWS change, intercomparison is from a variety of satellite-measured and simulated results. The plots
conducted by means of box plots of 4 GRACE- and 4 GLDAS-born results can quantify the skewness and kurtosis which are used to measure
(Fig. 2A–F). the asymmetry and peakedness of a real-valued random variable.
Firstly, it suggests that means of these GRACE- and GLDAS-born In most continents (Africa, Fig. 3A, Australia, Fig. 3C, Europe, Fig. 3D,
estimates of TWS change are closely consistent over the six major North America, Fig. 3E, and South America Fig. 3F), zero skewnesses
continents across the world during the last 10 years. Those means are identified in GRACE- and GLDAS-born results. In contrast, negative
are generally close to 0, which implies that TWS change in the past skewnesses are found in Australia (Fig. 3B). As an exception, the skew-
decade can be neglected although impacts of climate changes have ness in North America is slightly larger than zero. Generally, the skew-
been observed in many parts of globe. However, change in various ness is consistent with the average TWS anomaly shown by box plots
water resource components (e.g. glacial water, snow water, moisture (Fig. 2). Meanwhile, low kurtosises are identified in most estimates
water, ground water) is distinctive. except CLM (Fig. 3A, C, and F). Which means that the amplitude of TWS
With regard to the variances, the figure suggests that those of 4 change simulated by CLM is concentrated in small values, consistent
GRACE-based estimates (JPL, CNES, DMT, CU) are generally smaller with small variance in Fig. 2A, C, and F.
than those of GLDAS- based estimates (CLM, Mosaic, Noah, VIC). Among
all of the 4 GRACE-based estimates, variance of DMT is the smallest. 3.2. Spatial variability for various GRACE- and GLDAS-born estimates
Whereas, that of JPL is the largest while the variances of CLM and
Noah are smaller than those of Mosaic and VIC. Nevertheless, the A global grid-based view of spatial reliability for a range of satellite-
range of Noah estimates is more consistent than CLM estimates. Small observations and model-simulations is beneficial to identify the
ranges of CLM estimates are identified in Africa (Fig. 2A), Australia optimized estimates of the TWS anomaly in different parts of world.
(Fig. 2C), and South America (Fig. 2F), while large ranges of CLM esti- This section aims to examine the spatial correlation coefficients through
mates are found in Asia (Fig. 2B), Europe (Fig. 2D), and North America a global assessment.
(Fig. 2E). This suggests that Noah is most robust global land surface To examine the reliability of four TWS estimated from GRACE, an
model in GLDAS, which explains why it was widely used in different intercomparison is conducted with the GLDAS simulations. Fig. 4 shows
parts of the globe to simulate the TWS change (e.g. Globe, Syed et al., the spatial variations of the correlation coefficients between the monthly
2008, Eurasia, Velicogna et al., 2012). GRACE-born TWS anomaly and the average GLDAS-based TWS anomaly
 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37 35

80 A) CU 0.8
80 B) DMT 0.8

60 0.6 60 0.6

0.4 0.4
40 40
0.2 0.2
20 20
0 0

0 0
-0.2 -0.2

-20 -0.4 -20 -0.4

-0.6 -0.6
-40 -40
-0.8 -0.8

-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150

80 C) CNES 0.8
80 D) JPL 0.8

60 0.6 60 0.6

0.4 0.4
40 40
0.2 0.2
20 20
0 0

0 -0.2 0
-0.2

-0.4 -0.4
-20 -20

-0.6 -0.6
-40 -40
-0.8 -0.8

-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150

Fig. 4. The spatial pattern of the correlation coefficients between different GRACE-born TWS and the average TWS from the four GLDAS LSMs.

of four LSMs worldwide (Table 2). The GRACE-born TWS anomaly time these areas as shown in Fig. 1. Meanwhile, though the spatial perfor-
series show a positive correlation with GLDAS simulations in most places mance of the four GRACE-born TWS (JPL, CNES, DMT, CU) is very similar
across the world. However, there is a distinct negative correlation in (Fig. 4), JPL shows an excellent match with average GLDAS-born TWS in
Greenland, which could be attributed to the model's inability to simulate Northeastern Canada, the Scandinavian Peninsula in Northern Europe,
ice sheet flow and mass balance. In Northwestern Canada, Central Asia Western and Central Russia. This suggests that JPL is more competent
and Northern Africa, the correlation between GRACE-born and GLDAS- than CNES, DMT and CU in monitoring the TWS anomaly globally. Reduc-
born TWS change is poor. This could be the leading cause of poor agree- tion of postglacial rebound signal using background models may contrib-
ment between different GRACE-born and GLDAS-based TWS change in ute significantly to JPL's superiority over the other three GRACE products,

80 A) CLM 0.8
80 B) MOSAIC 0.8

60 0.6 60 0.6

0.4 0.4
40 40
0.2 0.2
20 20
0 0

0 -0.2 0 -0.2

-0.4 -0.4
-20 -20
-0.6 -0.6
-40 -40
-0.8 -0.8

-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150

80 C) Noah 0.8
80 D) VIC 0.8

60 0.6 60 0.6

0.4 0.4
40 40
0.2 0.2
20 20
0 0

0 -0.2 0 -0.2

-20 -0.4 -0.4

-20
-0.6 -0.6
-40 -40
-0.8 -0.8

-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150

Fig. 5. The spatial pattern of the correlation coefficients between the average GRACE-born TWS and TWS from the four GLDAS LSMs.
36 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37

Table 4
Advantages and disadvantages of GRACE- and GLDAS-born TWS anomalies worldwide.

Type Product Advantages Disadvantages

1. GRACE-born JPL-GRACE Tellus Better spatial correlation with GLDAS in Northeastern Canada,
the Scandinavian Peninsula in Northern Europe,
and Western and Central Russia
DEOSS DMT V 1b Small variance
CNES-GRGS V 2 Large variance
CU, Boulder Moderate variance
2. GLDAS-born CLM 2.0 Inconsistent variance worldwide
MOSAIC v3 Large variance
NOAH 2.7 Moderate and consistent variance worldwide
Better spatial correlation with GRACE in Northeastern Canada,
the Scandinavian Peninsula and Western and Central Russia
VIC 4.0.4 Large variance

especially in regions like the Scandinavian Peninsula and Northeastern the satellites (Güntner, 2008). Truncation and filtering also modify
Canada. the mean value of effective basin function and lead to a bias error of
Similarly, Fig. 5 presents the spatial variations of the correlation TWS change (Longuevergne et al., 2010). Bias and leakage can partially
coefficients between the mean GRACE-born TWS anomaly and the cancel one another over large areas where temporal water storage
different GLDAS-based TWS anomaly of the four LSMs worldwide. The variations are spatially homogeneous, thus filtered GRACE may agree
spatial pattern of correlation shown by Fig. 5 is highly consistent with well with unfiltered hydrological model simulations of water storage
that in Fig. 4. As indicated by the RMSE shown in Table 3, among the (Chen et al., 2005b). However, leakage effects tend to be reduced for
four estimates (CLM, Mosaic, Noah, VIC), Noah exhibits a better goodness areas near oceans where storage variations are generally smaller. And
of fit than CLM, Mosaic, and VIC. Many reports which used Noah (e.g. Syed the remaining bias effect may cause filtered GRACE data to significantly
et al., 2008; Leblanc et al., 2009; Grippa et al., 2011) also confirmed underestimate water storage variations in such areas, resulting in the
the point. poor correlation shown in Fig. 5 in regions such as the Scandinavian
Peninsula, Northeastern Canada, the western margin of South America
3.3. Underlying causes of inconsistency between GRACE- and GLDAS-born and the southern corner of Africa.
TWS estimates
4. Conclusion and recommendation
The inconsistency between GRACE- and GLDAS-born TWS changes
in Africa, Asia, and Europe (Fig. 1A, B and D) are marked by the rectangle This work strives to characterize the spatio-temporal variability of
boxes in Figs. 4 and 5. The difference between GRACE- and GLDAS-born different GRACE-born TWS over six major continents, and intercompare
TWS estimates can mainly be attributed to two main points. them with results from GLDAS simulations. The major points are
Firstly, TWS estimates from GRACE observations and GLDAS simula- summarized as following:
tions are not entirely equivalent. TWS from GRACE encompasses unsat-
urated and saturated soil water integrated over the entire soil column (1) In general, TWS anomaly time series estimated from four GRACE-
and therefore includes aquifers as well as surface water within rivers born products closely match with four GLDAS simulations in
and lakes. However, the total depth of soil moisture in CLM (10 layers), Australia and South America. The magnitude of GLDAS TWS anom-
Mosaic (3 layers), Noah (4 layers), and VIC (3 layers) models are 3.4 m, aly is larger than GRACE-born results in Europe and North America.
3.5 m, 2.0 m, and 1.9 m, respectively. Hence, the entire soil depth, GRACE-born and GLDAS-born TWS anomalies show poor agree-
in the land surface models employed here, is represented by a shallow ment in Africa and Asia.
soil reservoir, that is within 1.9 m (VIC) to 3.5 m (MOSAIC) beneath (2) The correlation between GRACE- and GLDAS-born TWS anoma-
the land surface. Groundwater contribution to TWS variations could lies is poor in Northeastern Canada, Central Asia (mainly in the
not be neglected, however is not included in the GLDAS simulations Tibetan Plateau and Xinjiang region), and North Africa (mainly
(Rodell et al., 2004b, 2009). Thus the poor coherence between GRACE the Sahara). However, different GRACE-born estimates are closely
and GLDAS results is partly due to GLDAS's inability to involve surface consistent with each other over such regions. Hence, poor correla-
water and groundwater components, especially in arid regions where tion between GRACE and GLDAS results may mainly arise from
groundwater can play a more important role to TWS change than soil LSMs' inability to involve groundwater storage variability and to
moisture (Figs. 1A and 4, North Africa; Figs. 1B and 4, Central Asia). On model the unique land surface processes in regions with extreme
the other hand, model simulations are mainly constrained by input cli- climatic conditions, such as the cold ice-cover regions (Canada
matic forcing data while GRACE is more capable of detecting TWS and the Himalayan region of Asia) or desert region (the Sahara).
change signals caused by other factors such as human activity (Voss Bias and leakage errors due to the GRACE processing procedure
et al., 2013). Hence, GRACE can capture stronger intra- and inter-annual may also contribute to poor correlation between GRACE and
variability than GLDAS, especially in the northern hemisphere where GLDAS.
human activity is more intensive. (3) With a comprehensive consideration on intercomparsion of the
Secondly, signal contamination due to leakage from surrounding advantages and disadvantages among GRACE- and GLDAS-born
regions and signal reduction due to GRACE processing procedures in- TWS anomalies, a summary is hereby recommended as a rapid
cluding truncation, destriping and smoothing may result in incoherence reference (Table 4) for scientists, end-users, and policy-makers
of GRACE TWS estimates compared to GLDAS simulations (Fig. 1B, Asia in the practices of global terrestrial water storage change research.
and D, Europe). Leakage contamination from mass signals outside the
area of interest is a main source of GRACE uncertainties (Swenson and References
Wahr, 2002). It is due to the fact that the gravity acceleration sensed
by the GRACE satellites is an integral effect of all masses both interior Bonan, G.B., et al., 2002. The land surface climatology of the community land model coupled
to the NCAR community climate model. Journal of Climate 15 (22), 3123–3149.
and exterior to the region of interest, with the contribution of individual Bruinsma, S., Lemoine, J.M., Biancale, R., Vales, N., 2010. CNES/GRGS 10-day gravity field
mass elements decreasing as a function of the squared distance to models (release 2) and their evaluation. Advances in Space Research 45 (4), 587–601.
 T. Yang et al. / Global and Planetary Change 109 (2013) 30–37 37

Chen, F., et al., 1996. Modeling of land surface evaporation by four schemes and com- Ramillien, G., et al., 2006. Time variations of the regional evapotranspiration
parison with FIFE observations. Journal of Geophysical Research Atmospheres 101 rate from Gravity Recovery and Climate Experiment (GRACE) satellite gravimetry.
(D3), 7251–7268. Water Resources Research 42 (W1040310).
Chen, J.L., Rodell, M., Wilson, C.R., Famiglietti, J.S., 2005a. Low degree spherical harmonic Ramillien, G., Famiglietti, J.S., Wahr, J., 2008. Detection of Continental hydrology and
influences on Gravity Recovery and Climate Experiment (GRACE) water storage glaciology signals from GRACE: a review. Surveys in Geophysics 29 (4–5), 361–374.
estimates. Geophysical Research Letters 32 (L1440514). Rodell, M., et al., 2004a. Basin scale estimates of evapotranspiration using GRACE and
Chen, J.L., Wilson, C.R., Famiglietti, J.S., Rodell, M., 2005b. Spatial sensitivity of the other observations. Geophysical Research Letters 31 (L2050420).
Gravity Recovery and Climate Experiment (GRACE) time-variable gravity observations. Rodell, M., et al., 2004b. The global land data assimilation system. Bulletin of The American
Journal of Geophysical Research Solid Earth 110 (B08408B8). Meteorological Society 85 (3), 381.
Cheng, M.K., Tapley, B.D., 2004. Variations in the Earth's oblateness during the past 28 years. Rodell, M., Velicogna, I., Famiglietti, J.S., 2009. Satellite-based estimates of groundwater
Journal of Geophysical Research Solid Earth 109 (B09402B9). depletion in India. Nature 460 (7258), 999-U80.
Dai, Y.J., et al., 2003. The common land model. Bulletin of The American Meteorological Rodell, M., McWilliams, E.B., Famiglietti, J.S., Beaudoing, H.K., Nigro, J., 2011. Estimating
Society 84 (8), 1013. evapotranspiration using an observation based terrestrial water budget. Hydrological
Ek, M.B., et al., 2003. Implementation of Noah land surface model advances in the National Processes 25 (26SI), 4082–4092.
Centers for Environmental Prediction operational mesoscale Eta model. Journal of Schmidt, R., et al., 2006. GRACE observations of changes in continental water storage.
Geophysical Research Atmospheres 108, 8851D22. Global and Planetary Change 50 (1–2), 112–126.
Famiglietti, J.S., 2004. Remote sensing of terrestrial water storage, soil moisture and Schmidt, R., et al., 2008. Periodic components of water storage changes from GRACE
surface waters. State of the Planet: Frontiers And Challenges In Geophysics 150, and global hydrology models. Journal of Geophysical Research Solid Earth 113
197–207. (B08419B8).
Fukuda, Y., et al., 2009. Monitoring groundwater variation by satellite and implications for Swenson, S., Wahr, J., 2002. Methods for inferring regional surface-mass anomalies from
in-situ gravity measurements. Science of The Total Environment 407 (9), 3173–3180. Gravity Recovery and Climate Experiment (GRACE) measurements of time-variable
Grippa, M., et al., 2011. Land water storage variability over West Africa estimated gravity. Journal of Geophysical Research Solid Earth 107 (2193B9).
by Gravity Recovery and Climate Experiment (GRACE) and land surface models. Swenson, S., Wahr, J., 2006. Post-processing removal of correlated errors in GRACE data.
Water Resources Research 47 (W05549). Geophysical Research Letters 33 (L084028).
Güntner, A., 2008. Improvement of global hydrological models using GRACE data. Swenson, S., Chambers, D., Wahr, J., 2008a. Estimating geocenter variations from a
Surveys in Geophysics 29 (4–5), 375–397. combination of GRACE and ocean model output. Journal of Geophysical Research
Klees, R., et al., 2008. The design of an optimal filter for monthly GRACE gravity models. Solid Earth 113 (B08410B8).
Geophysical Journal International 175 (2), 417–432. Swenson, S., Famiglietti, J., Basara, J., Wahr, J., 2008b. Estimating profile soil moisture
Koren, V., et al., 1999. A parameterization of snowpack and frozen ground intended and groundwater variations using GRACE and Oklahoma Mesonet soil moisture data.
for NCEP weather and climate models. Journal of Geophysical Research Atmospheres Water Resources Research 44 (W014131).
104 (D16), 19569–19585. Syed, T.H., Famiglietti, J.S., Rodell, M., Chen, J., Wilson, C.R., 2008. Analysis of terrestrial
Koster, R.D., Suarez, M.C.B.S.J., 1996. Energy and water balance calculations in the mosaic water storage changes from GRACE and GLDAS. Water Resources Research 44
LSM. NASA Technical Memorandum, 104606, 9 (76 pp.). (W024332).
Leblanc, M.J., Tregoning, P., Ramillien, G., Tweed, S.O., Fakes, A., 2009. Basin-scale, Tapley, B.D., Bettadpur, S., Ries, J.C., Thompson, P.F., Watkins, M.M., 2004. GRACE
integrated observations of the early 21st century multiyear drought in southeast measurements of mass variability in the Earth system. Science 305 (5683), 503–505.
Australia. Water Resources Research 45 (W04408). Tregoning, P., Lambeck, K., Ramillien, G., 2008. GRACE estimates of sea surface height
Liang, X., Lettenmaier, D.P., Wood, E.F., Burges, S.J., 1994. A simple hydrologically based anomalies in the Gulf of Carpentaria, Australia. Earth and Planetary Science Letters
model of land surface water and energy fluxes for general circulation models. Journal 271 (1–4), 241–244.
of Geophysical Research Atmospheres 99 (D7), 14415–14428. Velicogna, I., Tong, J., Zhang, T., Kimball, J.S., 2012. Increasing subsurface water storage
Liu, X., 2008. Global gravity field recovery from satellite-to-satellite tracking data with the in discontinuous permafrost areas of the Lena River basin, Eurasia, detected from
acceleration approach. Nederlandse Commissie voor Geodesie. GRACE. Geophysical Research Letters 39 (L09403).
Liu, X., et al., 2010. DEOS Mass Transport model (DMT-1) based on GRACE satellite data: Voss, K.A., et al., 2013. Groundwater depletion in the Middle East from GRACE with impli-
methodology and validation. Geophysical Journal International 181 (2), 769–788. cations for transboundary water management in the Tigris–Euphrates–Western Iran
Lo, M., Famiglietti, J.S., Yeh, P.J.F., Syed, T.H., 2010. Improving parameter estimation and region. Water Resources Research 49 (2), 904–914.
water table depth simulation in a land surface model using GRACE water storage Wahr, J., Molenaar, M., Bryan, F., 1998. Time variability of the Earth's gravity field:
and estimated base flow data. Water Resources Research 46 (W05517). hydrological and oceanic effects and their possible detection using GRACE. Journal of
Longuevergne, L., Scanlon, B.R., Wilson, C.R., 2010. GRACE hydrological estimates Geophysical Research Solid Earth 103 (B12), 30205–30229.
for small basins: evaluating processing approaches on the High Plains Aquifer, USA. Wahr, J., Swenson, S., Zlotnicki, V., Velicogna, I., 2004. Time-variable gravity from GRACE:
Water Resources Research 46 (W11517). first results. Geophysical Research Letters 31 (L1150111).
Paulson, A., Zhong, S., Wahr, J., 2007. Inference of mantle viscosity from GRACE and relative Yirdaw, S.Z., Snelgrove, K.R., Seglenieks, F.R., Agboma, C.O., Soulis, E.D., 2009. Assessment
sea level data. Geophysical Journal International 171 (2), 497–508. of the WATCLASS hydrological model result of the Mackenzie River basin using the
Ramillien, G., Cazenave, A., Brunau, O., 2004. Global time variations of hydrological signals GRACE satellite total water storage measurement. Hydrological Processes 23 (23),
from GRACE satellite gravimetry. Geophysical Journal International 158 (3), 813–826. 3391–3400.