Assessment of GRACE satellites for groundwater estimation in Australia

wastecypriotInternet and Web Development

Nov 10, 2013 (3 years and 9 months ago)

324 views

NATIONAL WATER COMMISSION


WATERLINES

i

Assessment of GRACE


satellites for groundwater


estimation in Australia

P. Tregoning, S. McClusky

Res
earch School of Earth Sciences

the Australian National University

A.I.J.M. van Dijk, R.S. Crosbie,



J.L. Peña
-
Arancibia

CSIRO Water for a Healthy Co
untry Flagship

Waterlines Report Series No

71,
February
2012










Waterlines

This paper is part of a series of works commissioned by the National Water Commission on
key water issues. This work has been undertaken by a consortium of scientists from T
he
Australian National University and CSIRO on behalf of the National Water Commission.




© Commonwealth of Australia
2012


This work is copyright.

Apart from any use as permitted under the
Copyright Act 1968
, no part may be reproduced by
any process with
out prior written permission.

Requests and enquiries concerning reproduction and rights should be addressed to the
Communications Director, National Water commission, 95 Northbourne Avenue, Canberra
ACT 2600 or
email
bookshop@nwc.gov.au
.


Online
/print
: ISBN:
978
-
1
-
921853
-
54
-
8

Assessment of GRACE satellites for groundwater estimation in Australia
,
February

2012

Authors: P

Tregoning, S

McClusky, A.I.J.M. van Dijk, R
S
Crosbie

and JL

Peña
-
Arancibia


Published by the Na
tional Water Commission

95 Northbourne Avenue

Canberra ACT 2600

Tel: 02 6102 6
000

Email:
enquiries@nwc.gov.au

Date of publication:
February
2012

Cover design by:
Angelink

Front cover image courtesy of
nasa.gov.au

An
appropriate citation for this report is:

Tregoning P et al, 2012,
Assessment of GRACE satellites for groundwater estimation in
Australia
, Waterlines report, National Water Commission, Canberra

Discla
i
mer

This paper is presented by the National Water C
ommission for the purpose of informing
discussion and does not necessarily reflect the views or opinions of the Commission.






NATIONAL WATER COMMISSION


WATERLINES

iv

Contents

Executive summary

................................
................................
................................
...................

ix

1. Introduction

................................
................................
................................
............................

1

1.1 The earth’s gravity field and the GRACE mission

................................
......................

1

1.2 Why use GRACE to monitor
groundwater?

................................
...............................

3

2. Review of existing studies applying GRACE to hydrology or groundwater estimation

.........

5

2.1 Hydrological studies u
sing GRACE products

................................
.............................

5

2.2 Review of applications of GRACE products for hydrological studies in
Australia

................................
................................
................................
............................

9

3. Assessment of the

available GRACE gravity fields

................................
.............................
12

3.1 GRACE products and their use

................................
................................
................
12

3.2 Comparison and validation of EWH solutions

................................
..........................
20

4. Interpreting GRACE water storage estimates

................................
................................
......
22

4.1 Introduction

................................
................................
................................
...............
22

4.
2 Review

................................
................................
................................
......................
23

4.3 Soil moisture storage estimation uncertainty due to rainfall estimation
error

................................
................................
................................
................................
26

4.4 Soil storage estimation unc
ertainty due to model error

................................
............
31

5. Derivation and Assessment of Groundwater Variations

................................
.......................
39

5.1 Point
-
scale groundwater level obser
vations

................................
.............................
39

5.2 Grid
-
scale groundwater level observations

................................
..............................
40

5.3 Time series comparison of groundwater level observations and
GRA
CE/AWRA

................................
................................
................................
...............
42

5.4

Assessment of the comparison in GWS between the GW data and
GRACE

................................
................................
................................
...........................
64

6. Known Errors and Estimates of Uncertainti
es of Remotely Sensed Groundwater

..............
67

6.1 Quantification of GRACE errors

................................
................................
...............
67

6.2 Quantification of modelled soil moisture errors

................................
........................
73

6.3 Groundwater uncertainty map for Australia

................................
..............................
73

7. Conclusions

................................
................................
................................
..........................
75

Bibliography

................................
................................
................................
..............................
77


Tables

Table 1: Average continental (including Tasmania) seasonal amplitude, trend and RMSD
when compared with AWRA
-
L for 2002

2010

................................
...............................
36

Table 2: Details on the grid cells chosen for further investigation (shown in Figure 19d)

........
42


Figures

Figure 1:
The GRACE space gravity mission (
nasa.gov.au
)

................................
......................

2

Figure 2: The earth’s gravity field, showing a) the latitudinal variation caused by the equatorial
bulge, b) a snapshot of geophysical processes

by computing anomalies at a single
epoch (i.e. residual signal about the mean value)
................................
..........................
13

Figure 3: Rate of change (in terms of EWH) for the CSR GRACE solutions (2002

2011) using
Gaussian filte
ring with radii from 0 km to 700 km

................................
..........................
16

Figure 4: Rate of change (in terms of EWH) for the period 2002

2011 derived from the GRGS
and CSR solutions using coefficient rates that pass an f
-
test w
ith statistical confidence
interval of 95%, 99%, 99.9% or 99.99%.

................................
................................
........
19

Figure 5: Rate of change in the Australian region (in terms of EWH/year) derived from several
different GRACE solutions spa
nning 2002

2010.
................................
..........................
21

NATIONAL WATER COMMISSION


WATERLINES

v

Figure 6: (a) Standard difference between GRACE and AWRA
-
L TWS anomalies (b) GRACE
water storage retrieval error estimates (c) Coefficient of correlation between GRACE
and
AWRA TWS anomalies (d) Colour composite showing the relative contribution of
the three signal components (seasonal cycle, eight
-
year trends, de
-
trended anomalies)
to the overall disagreement between GRACE and AWRA
-
L TWS (from Van Dijk et al.
2011)

................................
................................
................................
..............................
24

Figure 7: (a) Geographical distribution of active rain gauges (black dots) during 1998

2008
used in generating precipitation forcing data (b) Areas with >20 unreliable data (in blue)
during 1911

20
10 (after BoM 2011)

................................
................................
..............
26

Figure 8: Summary statistics for the three member ensemble precipitation data (SILO,
BAWAP and LSBLEND) for 1998

2008.

................................
................................
.......
28

Figure 9: Comparison of precipitation monthly correlation (r) and root mean squared
difference (RMSD) for all months in 1998

2008.

................................
...........................
29

Figure 10: Comparison of AWRA
-
L modelled
soil moisture storage (SMS) trends with different
precipitation forcing for 1998

2008.

................................
................................
...............
30

Figure 11: Comparison of AWRA modelled soil moisture correlation with different precipitation
forcing
for 1998

2008.

................................
................................
................................
...
31

Figure 12: Modelled soil moisture seasonal amplitude for 2002

2010

................................
....
33

Figure 13: Modelled SMS trend for 2002

2
010

................................
................................
.......
34

Figure 14: Root mean square difference (RMSD) in soil moisture storage (SMS) anomalies for
2002

2010, between the four GLDAS models and AWRA
-
L

................................
........
35

Figure 15: Averaged root mean square difference between soil moisture storage (SMS)
estimates from the four GLDAS models and AWRA

................................
......................
35

Figure 16: (a) Monthly uncertainty

time series (in the form of standard deviation) from the
AWRA
-
L and GLDAS models evaluated in the Canning basin near Broome (E122.5º,
S17.5º) (b) Ensemble SMS change (blue line and dots) showing standard deviation
bars (grey) (c) Time series of SMS change

from AWRA
-
L and GLDAS

........................
37

Figure 17: (a) Monthly uncertainty time series (in the form of standard deviation) from the
AWRA
-
L and GLDAS models evaluated in the Condamine basin (E148.5º, S27.5º)

(b)
Ensemble SMS change (blue line and dots) showing error bars (grey) (c) Time series of
SMS change from the AWRA
-
L and GLDAS

................................
................................
.
38

Figure 18: Trend in groundwater level at each monitoring bore

that has at least five
measurements spread over at least two years in the period 1/7/2002 to 30/6/2010

.....
39

Figure 19: Trends in groundwater level from monitoring bores aggregated to a grid sca
le

.....
41

Figure 20: Method used to create a time series of EWH from multiple observation bores within
a grid cell

................................
................................
................................
........................
44

Figure 21: Sou
rces of uncertainty in the calculation of the combined time series at the Lachlan
grid cell

................................
................................
................................
...........................
45

Figure 22: Differences in the combined time series at Lachlan of assuming different values of
spe
cific yield (note the different y
-
axis scales)

................................
...............................
45

Figure 23: Surface geology of the grid cell near Broome with the location and trend in the
observation bores

................................
................................
................................
...........
46

Figure 24: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell near Broome

................................
........................
47

Figure 25: Surface geo
logy of the grid cell near Telfer with the location and trend in the
observation bores

................................
................................
................................
...........
48

Figure 26: A comparison of the change in groundwater storage derived from observation
bores and GRACE
from the grid cell near Telfer

................................
...........................
48

Figure 27: Surface geology of the grid cell in the Daly West with the location and trend in the
observation bores

................................
................................
................................
...........
49

Figure 28: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell in the Daly West
................................
...................
50

Figure 29: Surface geology o
f the grid cell in the Daly East with the location and trend in the
observation bores

................................
................................
................................
...........
51

NATIONAL WATER COMMISSION


WATERLINES

vi

Figure 30: A comparison of the change in groundwater storage derived from observation
bores and GRACE f
rom the grid cell in the Daly East
................................
....................
51

Figure 31: Surface geology of the grid cell in the Fitzroy East with the location and trend in the
observation bores

................................
................................
................................
...........
52

Figure 32: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell in the Fitzroy East

................................
................
53

Figure 33: Surface

geology of the grid cell in the Fitzroy West with the location and trend in
the observation bores

................................
................................
................................
.....
54

Figure 34: A comparison of the change in groundwater storage derived from observation
bore
s and GRACE from the grid cell in the Fitzroy West

................................
...............
54

Figure 35: Surface geology of the grid cell in the Brisbane Catchment with the location and
trend in the observation bores

................................
................................
........................
55

Figure 36: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell in the Brisbane Catchment

................................
..
56

Figure 37: Surface geology of the grid cell in the Condamine East with the location and trend
in the observation bores

................................
................................
................................
.
57

Figure 38: A comparison of the change in groundwater storage

derived from observation
bores and GRACE from the grid cell in the Condamine East

................................
........
57

Figure 39: Surface geology of the grid cell in the Condamine West with the location and trend
in the obser
vation bores

................................
................................
................................
.
58

Figure 40: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell in the Condamine West

................................
.......
59

Figure 41: Surface geology of the grid cell in the lower Lachlan with the location and trend in
the observation bores

................................
................................
................................
.....
60

Figure 42: A comparison of the change

in groundwater storage derived from observation
bores and GRACE from the grid cell in the lower Lachlan

................................
.............
60

Figure 43: Surface geology of the grid cell near Renmark with the location and trend

in the
observation bores

................................
................................
................................
...........
61

Figure 44: A comparison of the change in groundwater storage derived from observation
bores and GRACE from the grid cell near Renmark

................................
......................
62

Figure 45: Surface geology of the grid cell near Shepparton with the location and trend in the
observation bores

................................
................................
................................
...........
63

Figure 46: A comparison of the change
in groundwater storage derived from observation
bores and GRACE from the grid cell near Shepparton

................................
..................
63

Figure 47: Scatter plot of the GWS derived from GW data and GRACE (GRGS solution minus
AWR
A
-
L)

................................
................................
................................
........................
65

Figure 48: a) Time series of EWH change (and formal uncertainties) from the GRGS GRACE
solutions evaluated at location E122º, S22º, b) Time series of the formal uncertainties
themselve
s, c) Histogram of the formal uncertainties

................................
....................
68

Figure 49: Histogram of uncertainties in GRACE EWH at 145ºE for latitudes 5ºS, 45ºS and
85ºS

................................
................................
................................
................................
69

Figure 50: Amplitude of S
2

ocean tide errors in GRACE solutions, aliased to 161
-
day period
signal in EWH time series

................................
................................
..............................
70

Figure 51: Amplitude of the annual variations in GRACE so
lutions

................................
.........
71

Figure 52: Standard deviation (of a single observation about the mean) of the MOG2D
-
G
barotropic ocean model

................................
................................
................................
..
72

Figur
e 53: Estimated overall uncertainty in SMS estimates.

................................
....................
73

Figure 54: Map showing likely uncertainties in groundwater estimates derived from a
combination of GRACE TWS and SMS

................................
................................
.........
74


NATIONAL WATER COMMISSION


WATERLINES

vii

Abbreviations and acronyms

AARR

Accumulated Annual Rainfall Record

AWRA

Australian Water Resources Assessment system

AWRA
-
L

AWRA Landscape hydrology model

BAWAP

Bureau of Meteorology Australian Water Availability Project

BoM

Bureau o
f Meteorology

C
20

Degree 2, Order 0 spherical harmonic coefficient that describes the equatorial
bulge of the
e
arth

CSR

Center for Space Research, University of Texas at Austin, USA

EWH

Equivalent Water Height

GFZ

GeoForschungsZentrum (German Research
Centre for Geosciences)

GLDAS

Global
Land Data Assimilation System

GRACE


Gravity Recovery and Climate Experiment

GRGS

Groupe de Recherche de Géodésie Spatiale (Space Geodesy Research
Group, France)

GWL

groundwater level

GWS

groundwater storage

IOD

I
ndian Ocean dipole

ITG

Institute of Geodesy and Geoinformation, University of Bonn, Germany

JPL

NASA Jet Propulsion Laboratory

LAGEOS

Laser Geodynamics Satellites

LSBLEND

blended satellite
-
gauge precipitation estimates
(Li and Shao

2010
)

MDB

Murray

Da
rling Basin

MOG2D
-
G

2
-
dimensional gravity waves barotropic model
of Carrère and Lyard (
2003
)

NATIONAL WATER COMMISSION


WATERLINES

viii

NOAH

N: National Centers for Environmental Prediction; O: Oregon State University

(Department of Atmospheric Sciences); A: Air force; H: Hydrologic Research

Lab

RMSD

Root Mean Square Difference

SILO

Specialised Information for Land Owners spatial precipitation estimates

SMS

Soil Moisture Storage

SWS

Surface Water Storage

TWS

Total Water Storage

WIRADA

Water
Information Research and Development Alliance




NATIONAL WATER COMMISSION


WATERLINES

ix

E
xecutive summary

Groundwater management and GRACE

Groundwater is an important resource for many water users in Australia. Water managers
need information on the character, dynamics and current status of groundwater resources to
inform the planning and adju
stment of groundwater management regimes. Ongoing
challenges in groundwater management include the expense and scarcity of groundwater
mapping and monitoring, the high spatial variability in groundwater system characteristics and
the complexity
of
groundwa
ter storage dynamics. This combination means that local
measurements cannot be interpreted over larger areas without introducing large uncertainty.

The Gravity Recovery and Climate Experiment (GRACE) space gravity mission was launched
in 2002 with a plann
ed 5
-
year lifetime. The
mission
, a scientific and
technical
success, is still
functioning today. GRACE mass variation estimates over Australia quantify changes in total
water storage expressed as an Equivalent Water Height (EWH). Estimating groundwater
cha
nges then
requires
separating the total water storage changes into the components of
surface water, soil moisture
, biomass

and groundwater.


The reliability and accuracy of GRACE
-
derived groundwater storage changes depends upon
both the GRACE total water s
torage estimates and the soil moisture content estimates being
accurate and containing no systematic biases or trends. To estimate reliable large
-
scale
groundwater storage changes from
discrete

measurements in monitoring bores
,

the bore level
observations
must be representative of the groundwater variations at larger scales and the
specific yield (or percentage of water per volume of subsurface material) must be known
accurately to enable the conversion from groundwater levels to groundwater volumes.
These
requirements are critical to the resulting accuracy of each technique, and errors will degrade
the agreement in the comparison of groundwater estimates from the two techniques.

Errors in groundwater
storage

estimates derived from this process will be the
summation of
the errors in the GRACE total water storage changes, the modelled soil moisture values and
the surface water estimates.
Studies to date
,

and analysis in this report
,

show that the
greatest uncertainty originates from the separation between soi
l moisture and groundwater
:

in
other words, separating storage in the unsaturated and saturated zones.
Careful
consideration of the assumptions and processes involved
can
le
a
d to the generation of a
map
that shows the accumulated uncertainty in groundwater

storage estimates over Australia
derived from remote sensing observations
.

The two most commonly used GRACE products are those of the Centre
for

Space Research
(CSR) at the University of Texas, Austin, and the French Groupe de Recherche de Géodésie
Spatia
le (GRGS). The CSR fields must undergo filtering and scaling procedures before being
used to estimate water mass changes. The GRGS solutions undergo a regularisation during
the generation of the products
, and

hence, can be used directly without subsequent
filtering.
We assessed the likely errors in the rate fields of both CSR and GRGS solutions and then
focused our error analysis on just the GRGS solutions
,

because they seem to provide the
best agreement with soil moisture and groundwater bore information a
cross Australia.

Objective

The
goal
of this study is

to
evaluate the potential utility of GRACE observations for deriving
estimates of groundwater storage changes
.
P
recondition
s

of such a use
are

(1) that estimates
of groundwater storage can reliably be de
rived from GRACE for at least some parts of
Australia, (2) that the estimation accuracy is sufficient (that is, the uncertainty sufficiently
small) to be useful for management, and (3) that
estimates can be reconciled with
measurements in monitoring
bore
s,

where available.

In this report, we
independently
quantif
ied

the likely level and spatial and temporal variations
of error in each of the above assumptions
. We also attempted to reconcile the two
NATIONAL WATER COMMISSION


WATERLINES

x

independent estimates of groundwater storage variations for

12 1°x1° grid cells where
groundwater level variations from a reasonably large number of bores could be obtained.

Results

GRACE total water storage estimates around Australia are affected by errors in modelled
ocean mass movement

both tidal and non
-
tidal
. This is most problematic around the coast,
northern Australia and in the region of Gulf St Vincent. The formal uncertainties of the
GRACE estimates at any epoch increase from
approximately
21 mm EWH in Tasmania to
26

mm EWH in Cape York, caused by change
s in the spatial separation of the GRACE
satellite ground tracks.

Errors in the precipitation models used to force the hydrological models induce variations in
soil moisture estimates of more than
30 mm/month where rainfall is high and seasonal, spatial
ra
infall gradients are high and the density of gauges low. Most of this error is random, but we
found systematic differences in linear soil moisture trends of >5 mm per year. Differences in
model assumptions, structure and parameters cause large systematic d
ifferences in soil
moisture estimates between models, with the greatest monthly differences (>20 mm) in
regions with high rainfall and a strong seasonality. Strong differences in linear trend
s

(>10

mm per year) were found in northern Queensland and Tasmani
a, while differences in the
seasonal amplitude in soil moisture storage dominated elsewhere. The lack of accurate
knowledge about the maximum capacity of the soil to store and retain water affects, in
particular, the estimated seasonal amplitude in the mod
els. This is caused by uncertainty in
depth to groundwater, active root zone depth, and soil hydraulic properties.

The error in the specific yield is very difficult to quantify and acts as a scale factor in the
conversion
from groundwater level

in borehole
s to changes in EWH. The distribution of
monitored groundwater boreholes is not homogeneous and
was commonly biased towards
certain areas or groundwater systems. Accounting for this sampling problem would require a
good understanding
of
local hydrogeology
and the characteristics of the monitoring bores.

The comparison
between

groundwater storage changes derived from GRACE and model soil
moisture
,

and those derived from bore data produced mixed results. For a few regions, the
direction of bore levels and GRA
CE
-
inferred groundwater storage was opposite. In some
cases the bore estimates were also opposite to those estimated from rainfall patterns directly
,

while in other cases there was no consistency in linear trend between individual bores within
the grid cel
l, with both increasing and decreasing trends observed. Generally, for regions with
a large number of bores (Lachlan, Renmark, Shepparton)
,

there was better agreement
between GRACE
-

and bore
-
derived water storage.

In cases with strong seasonality in GRACE
water storage (e.g. northern Australia)
,

the
modelled soil moisture accounted for most of that variability, whereas the bore estimates
suggested that groundwater, too, had a strong seasonal cycle. This implies that shallow
groundwater changes have been inc
luded in the model parameterisation of soil storage
capacity.

In summary, our results indicate that three sources of uncertainty prevent us from making a
direct comparison between the two methods of groundwater storage estimation, namel
y,

(1)
hydrological

model assumptions required to
estimate soil moisture dynamics,

(2) the scarcity
and biased positioning
of groundwater monitoring bores,

and (3) specific yield assumptions
that need to be made to translate groundwater level
s

into storage.

Recommendations

For the few regions with suffic
i
ent bores to allow a good comparison (e.g. Shepparton,
Renmark), we found
arguably reasonable agreement
1

in derived groundwater storage



1

The agreement (or otherwise) between the two

techniques is detailed in Chapter 5.

NATIONAL WATER COMMISSION


WATERLINES

xi

estimates. Nonetheless
,

some distinct differences were found and these lead to the follo
wing
recommendations.

1.

A major source of uncertainty in deriving groundwater dynamics from GRACE is the need
to subtract estimated soil moisture storage. For most of the 12 regions investigated, it was
not
possible to
reliably
infer seasonal cycles in groun
dwater
storage
from
GRACE
because of
uncertainty in the seasonal cycle of water storage in the unsaturated zone
.

This
uncertainty
can be
reduced

by improving the soil moisture modelling using better
spatial information on depth to groundwater, subsurface h
ydraulic properties and
vegetation rooting depth, and improved representation of groundwater discharge
processes in the hydrological model used. This requires a combination of field
hydrological process knowledge and a sufficient number of observations of
groundwater
and soil water behavio
u
r in space and time. Such models may exist for certain regions.
On a continental scale, CSIRO and the Bureau of Meteorology are currently improving the
Australian Water Resources Assessment system along these lines.

Recom
mendation 1:

To
interpret

GRACE observations of groundwater variations, it is first
necessary to identify or develop hydrological models that cover a sufficiently large area
and which are known to describe saturated and unsaturated dynamics (and their
coup
ling) reliably.

2.

GRACE can add an overall constraint on
a sufficiently reliable
model, by providing the
total monthly water storage changes, each with an accuracy of
approximately
25 mm
EWH.

Moreover, there is no reason to assume the presence of systematic
errors such as
long
-
term drift

in the monthly GRACE solutions.

T
herefore, a particular strength of the
GRACE data is in providing valuable information on inter
-
annual changes in water
storage over large areas. Methods are required to constrain finer
-
resolu
tion models with
these observations.

Recommendation

2
:

Research needs to be conducted into how to assimilate GRACE total
water storage into hydrological models in Australia.

3.

There is potential for GRACE

observations
to
help improve the translation of groun
dwater
level changes measured in bores into
groundwater volumes. Comparisons of the two
independent groundwater estimates could be used to derive specific yield values on
broad scales,
and these

could be used to
extrapolate
estimates derived
locally
from
b
ore
pumping tests.

Recommendation

3
:

A study should be undertaken of the feasibility and accuracy of
specific yield estimates from the comparison of GRACE, soil moisture and groundwater
levels from borehole measurements.

4.

T
he utility
of
GRACE
-
derived water

storage estimates and the ability to reconcile these
with bore measurements
is limited
by
the
coarse resolution of GRACE
TWS estimates.

Recommendation
4
:

Improvements in the spatial resolution of GRACE products, tailored
for the Australian hydrological co
mmunity, need to be made in order to make the GRACE
products more relevant for the Australian groundwater community.


NATIONAL WATER COMMISSION


WATERLINES

1

1. Introduction

Knowledge and understanding of groundwater systems is complicated by the fact that it is
difficult and expensive to make

observations of groundwater levels. Traditional methods
involve the drilling and monitoring of groundwater bores, yet such approaches provide only
discrete sampling and limited knowledge on catchment and/or basin scales. Nonetheless, the
observations of g
roundwater levels at such bores have provided the only knowledge on the
changes in water resources in groundwater systems.

With the launch of the Gravity Recovery and Climate Experiment (GRACE) mission in 2002, a
new capability to observe total water stora
ge (TWS) at broad spatial scales became available.
GRACE detects the integrated change in mass of all components of the hydrological cycle,
including groundwater, soil moisture and surface storage. Thus, there is the possibility of
deriving groundwater var
iation estimates if the hydrology signals other than groundwater can
be subtr
acted from GRACE TWS estimates.


This report investigates the potential of using the GRACE space gravity mission, in
conjunction with modelling of soil moisture storage (SMS), to
derive estimates of broad
-
scale
groundwater changes. In this chapter we describe the GRACE mission and the potential it
offers for monitoring groundwater. Chapter 2 describes some of the pioneering hydrological
studies conducted using GRACE observations as

well as applications in the Australian region.
Chapters 3 assesses some of the available GRACE products

and

explains how they should
be used and their known limitations. In Chapter 4 we describe
methods to estimate the
influence of terrestrial mass change
s other than groundwater, and in particular soil moisture
storage,
and
uncertainties in accounting for these influences
.

In Chapter 5 we derive
estimates of groundwater storage change
for a number of regions in
Australia

where we can
compar
e

these
with gro
undwater
storage change estimates derived
from
groundwater bore
measurements
. An assessment of the known biases in the GRACE and
soil moisture
estimation
, including their spatial variability, is used to generate a groundwater uncertainty
map

in
Chapter
6
.
This can be used to assess where remotely sensed groundwater estimates
are likely to be more reliable. Conclusions and recommendations are made in Chapter 7.

1.1 The
e
arth’s gravity field and the GRACE
mission

The Gravity Recovery and Climate Experiment (
GRACE) space gravity mission is a joint
mission by NASA and the German Deutsche Forschungsanstalt für Luft und Raumfahrt (DLR)
mission. Launched in 2002 with a planned 5
-
year lifetime
2
, the single
e
arth observing mission
has brought together a number of di
fferent disciplines, providing information at broad spatial
scales. There are well over 100 scientific publications each year that depend on GRACE data.
GRACE data ha
s

been used to study geophysical processes on
e
arth including earthquake
deformation, melt
ing of continental ice

and

oceanic and hydrologic processes. Temporal
estimates (monthly or 10
-
daily snapshots) of the
e
arth’s gravity field are publicly available as
Level
-
2 products from the GRACE mission (described in Chapter 3). The GRACE mission is
ex
pected to survive until (at best) 2014, while the replacement GRACE Follow
-
o
n mission is
not scheduled for launch until 2017.

Gravity

is much weaker than other basic natural forces such

as strong and weak nuclear
interaction and electromagnetism
.

B
ut
grav
ity’s effects are ubiquitous and dramatic.
It

plays a
significant role in controlling

everything from the
e
arth’s tides to the expan
sion of the

u
niverse.




2

T
he mission is still functioning today, although many components are now in critical status and
batteries are starting to fail. Unforseen failure of this mission would result in no such space
-
based
gravity observatio
ns of the Earth being available until the launch of the GRACE Follow
-
o
n mission,
currently scheduled for 2017.

NATIONAL WATER COMMISSION


WATERLINES

2

Gravity

is
a natural phenomenon

by which physical bodies attract

with a
force

that is

proportional to

their mass.

M
ass refers to the amount of matter
contained within a given
space

and is directly related

to the density of a
material.
As an

example, a
volume

fi
lled with
more dense material
, like
rock
, has more mass than that same
volume

filled with water.

Since

mass and density are directly related, there is also a direct relationship between density and
gravity. An increase in density results in an increase in mass, and an increase in mass results
in an increase in the gravitational force exerted by
the

v
olume
.
Mass

fluctuations on
the
surface of the
e
arth, and within the
e
arth’s interior
,

therefore
,

cause

va
riations in the gravity
field.
The branch of science
that deals
with obtaining precise measurements of the
e
arth,
including its geometric shape and gr
avitational field
, is known as geodesy.

Since the first artificial
e
arth satellite was launched in 1957 (Sputnik), geodesists have used
observations of and from satellites to improve our knowledge

of t
he
e
arth's
gravity field
.
While
t
hese

early gravity

mea
surements described

the large
-
scale features

of
e
arth's gravitational
field they

could not

resolve
the
finer
-
scale features or accurately describe the small month
-
to
-
month variations associated with
mass redistributions on and within the
e
arth.
To learn mo
re
about the
e
arths’ gravity, in particular its time variable nature, the twin GRACE satellites were
laun
ched in

2002 with the primary goal to
precisely
measure the changing gravity field of the
e
arth.

Figure
1
: The GRACE
space

gr
avity mission (
nasa.gov.au
)


GRACE is the first
e
arth
-
monitoring mission in the history of space flight whose key
measurement is not derived from electromagnetic waves either reflected off, emitted by, or
transmitted through the
e
arth's surface and/or atm
osphere. Instead, the mission uses a
microwave ranging system to
accurately
measure changes in the speed and distance
between two identical spacecraft flying in a polar orbit about 220 kilometers a
part, 500
kilometers

above the
e
arth (Figure 1). The rangin
g system is sensitive enough to detect
separation changes as small as 10 micrometres (approximately one
-
tenth the width of a
human hair) over
a distance of 220 kilometers.
(
'
GRACE Launch Press Kit
'

http://grace.jpl.nasa.gov/files/GRACE_Press_Kit.pdf
)
.

Circ
ling the globe every 90 minutes, the twin GRACE satellites sense infinitesimal variations
in
e
arth's gravitational field. When the first satellite
approaches a region of

stronger gravity,
called
a
'
gravity anomaly
'
, it is
accelerated towards it
. This cause
s the distance between the
two satellites to increase. T
he first spacecraft lingers
over
the anomaly because
it is
decelerated by it
.

M
eanwhile the fo
llowing spacecraft is accelerated and will catch up to the
NATIONAL WATER COMMISSION


WATERLINES

3

first satellite, thus decreasing the distance b
etween them. The first satellite will continue past
the anomaly while the second is still retarded by it and so the distance between the satellites
increases.

This continuous change in distance between the satellites is caused directly by the highs and
lo
ws of the gravity field.
By constantly measuring the changing distance between the two
satellites and combining
th
at

data with precise measurements
of the GRACE satellites
'

absolute positions
from Global Positioning System (GPS) instruments

onboard, we

can

construct a detailed map of
e
arth's gravity as a function of time.

The two satellites constantly maintain a two
-
way
K/Ka
-
band
micr
owave
-
ranging link
3

between
them
.
Precise a
ccelerometer
s

located at the center of mass of each satellite are used
to
distingu
ish

(
and correct for
)

accelerations caused by non
-
gravitational sources such as
atmospheric drag, solar radiation and satellite thruster firings
. All of this information is
downloaded to ground stations. To
maintain

correct
baseline
separation

and
proper
o
rientation of each spacecraft, the satellites

use star cameras, magnetometers, and GPS
observations
. The GRACE vehicles also have optical corner reflectors to enable laser ranging
from ground stations, bridging the range between spacecraft positions and Do
ppler ranges
.

(
'
GRACE Mission Overview
'

http://www.csr.utexas.edu/grace/overview.html
)
.

Visit
http://www.csr.utexas.edu/grace

for a
dditional inf
ormation about the Gravity Recovery
and Clim
ate Experiment.


1.2 Why use GRACE to monitor groundwater?

While in situ hydrologic measurements provide discrete sampling of soil, ground and surface
water, GRACE gravity observations provide a unique quantitat
ive measurement of TWS
anomalies that are not available to hydrologists by any other practical means. GRACE

gives

hydrologists

the ability

to close the terrestrial water storage budget by providing a quantitative
estimate of total integrated water mass cha
nge over time. With nearly
10

years of GRACE
observations, long
-
term trends in terrestrial TWS can now be reliably assessed and compared
with hydrological models and standard drought indices.


The combination of remotely sensed total water storage changes
from GRACE
and

SMS
modelling and surface water estimates
,

offers the possibility to estimate groundwater changes
without the costly effort of drilling and instrumenting discrete groundwater bores. If shown to
be sufficiently accurate, this could provide a
totally new spatial and temporal dataset for
groundwater monitoring, enabling observation of all the aspects of the hydrological cycle.

U
ntil recently
,

o
ne of t
he major factors limiting the usefulness of GRACE estimates in
hydrological models has been its
relatively low native spatial resolution (about 350 km).
Recent progress, however, has been made in reducing this spatial resolution by customising
GRACE analysis for particular regions, catchments and drainage basins
,

and has enabled
GRACE to provide valu
able information on fine
-
scale integrated mass redistribution. For
examp
le, the work of Wouters et al. (
2008
)

showed t
hat using a forward modelling
('fingerprint'
) approach allowed for better spatial resolution of time variable masses changes
in Greenland
to be derived than could be achieved using the original spherical harmonics
directly. Simila
rly, the paper by Kurtenbach et

al.
(
2009
)

that applied a Kalman filter approach
to steer the spherical harmonic solutions was able to resolve spatially variable un
loading rates
over different regions of the Greenland ice sheet. More recently,
Longuevergne et al. (2010)

developed
a
mass concentration algorithm
,

called spatiospectral localisation
,

to study the US
High Plains aquifer, which optimises drainage basin sha
pe descriptions, taking into account
GRACE’s limited spatial resolution

and noise characteristics
. This method appears to be



3

The K/Ka
-
band microwave link is the inter
-
satellite range measuring system that provides the
information that makes the GRACE mission unique at

this time in being able to detect accurately the
temporal changes in the Earth’s gravity field. Changes in the separation distance of the two spacecraft
are related to the strength of the gravity field, which changes with both spatial location and time.

NATIONAL WATER COMMISSION


WATERLINES

4

particularly suited to retrieval of basin

scale TWS variations and is effective for basins as
small as 200,000 km
2

(
e.g. Longueverg
ne et al. 2010; Luthcke et al.

2006
)
.

Since launch in 2002, GRACE has been proven reliable, and offers a great potential for

water
storage budget closure on basin to regional scale
(
S
wenson et al. 2006; Yeh et al.

2006
)
.
GRACE data
is

available for virtually all river basins and can be used to estimate water
storage change in the thin lay
er at the surface of the
e
arth (
Brunn
er
et al. 2006; Swenson et
al.

2006
)

with unprecedented accuracy (
Tapley et al.

2005
)
. GRACE is promising because no
other global network exists of hydrological observations with temporal and spatial resolutions
necessary to characterise storage on

regional t
o continental scale (
Swenson et a
l. 2006;
Klees et al. 2006; Chen et al.

2007
)
.

NATIONAL WATER COMMISSION


WATERLINES

5

2. Review of existing studies applying
GRACE to hydrology or groundwater
estimation

In this chapter we assess how GRACE data ha
s

been used to study hydrological processes.
We
begin in Section 2.1 with simulation studies that were used prior to the launch of the
GRACE satellites to demonstrate the likely capability of the mission and capacity to estimate
signals associated with groundwater, surface water and soil moisture. Some
of the extreme
climate events over the past decade are described as seen by GRACE. We then look in detail
at some of the groundwater studies that have been undertaken and at some attempts to
validate
,

through in situ observations
,

the estimates of terrestr
ial water storage change from
GRACE. In Section 2.2 we focus on the applications of GRACE data to studies of Australian
hydrology.

2.1 Hydrological studies using GRACE products

Prelaunch assessments of the anticipated results from the GRACE mission showed

that
monthly, seasonal and annual changes in water storage within drainage basins should be
detectable in basins of approximately 200

000 km
2

(Rodell and Famiglietti

1999
)
. The primary
controls on the detectability of the signals were thought to be driven

by the GRACE
instrumental errors, atmospheric modelling errors in the region of the drainage basin and the
magnitude of the water storage changes themselves.

The first published results using data from the GRACE mission showed significant
improvement in
the accuracy with which the
e
arth’s g
ravity field could be measured (Tapley et
al.

2004
)

and yielded the first estimates of the amplitude of annual variations in the global
hydrological cycle. However, the results were about 40 times worse than the predict
ed
accuracy from prelaunc
h simulations (Wahr et al.

2004
)
. Significant errors in a north

south
striping pattern were evident in the solutions, completely masking the hydrological and
oceanic signals that were being sought. The stripes were found to be rela
ted to unidentified
errors in the reduction of the raw observations and filtering techniques were e
mployed to
reduce these errors (Tapley et al. 2004; Wahr et al.

2004
)
.

Subsequently, hundreds of studies using GRACE have been undertaken to quantify
hydrolo
gic, oceanic and climatic changes on the
e
arth. These include the estimation of snow
mass (Frappart et al.

2006
)
, the derivation of steric sea level variations
4

(e.g. Lombard et al.

2007
)
, the seasonal exchange of water

between oceans and continents (
Chamb
ers et al.
2004
)
,
and
glacial isostatic adjustment
5

(e.g. Tamisiea et al. 2007; Tregon
ing et al. 2009a; Wu
et al. 2010
; Ivins et al.

2011
).

In this chapter, we review some of the original studies that demonstrated the capabilities of
the GRACE mission. We
also provide examples of recent studies that show how improved
analysis techniques have led to greater accuracy in the estimation of mass changes. We
divide the discussion into studies of TWS changes, examples of extreme climate events
(droughts, floods
,

e
tc
.
), quantification of only groundwater variations and, finally, the validation
of GRACE estimates.




4

Steric sea level variations are the increases or decreases of sea surface heights through the
combination of thermal expansion/contraction and density changes related to salinity variations.

5

Glacial isostatic adjustment is the return to a state of isost
atic (or buoyancy) equilibrium of the Earth’s
crust as a result of changes in the mass of the ice sheets on the continents since the Last Glacial
Maximum about 20

000 years ago.

NATIONAL WATER COMMISSION


WATERLINES

6

2.1.1 Total water storage (TWS) studies

Despite its importance, TWS at regional and continent
al scales remains poorly known
(Ramillien et al.

2008
)
, largel
y because of a lack of systematic
and comprehensive
observations (
Lettenmaier and Famiglietti 2006
)
. The prelaunch s
tudy of Rodell and
Famiglietti (
1999
)

investigated the feasibility of detecting monthly, seasonal and trend signals
in drainage basins of di
fferent spatial scales, given a likely range of errors of the original
GRACE observations. They found that:



monthly changes in TWS should be detectable 50

91% of the time in 15 of 17 basins
larger than 200

000 km2



seasonal signals should be detectable 50

100% of the time in 17 of 18 basins larger than
184

000 km
2



annual variations should be detectable in 13 of 17 basins larger than 200

000 km
2
.

When launched, the GRACE
s
cience
t
eam encountered difficulties in achieving the expected
level of accuracy and it

took nearly two years before the data w
as

released publicly. The
publication of
Tapley et al. (
2004
)

contains the first published results and shows clearly the
annual variations globally and, in particular, over the Amazon/Orinoco river systems. They
also

provided the first attempts at estimating temporal trends, although the time series used
contains only 14 months of GRACE data.

Rodell et al.
(
2004
)

found that the GRACE TWS estimates lay roughly between estimates
derived from a water balance model and th
e Global Land Data Ass
imilation System (GLDAS)
model (
Rodell et al. 2004
)

drivi
ng the NOAH land surface model (Ek et al.

2003
)
. They also
found that the spatial scaling applied to the GRACE data affected the amplitude of the
variations in the GRACE estimat
es (see Section 3.1.2 for a detailed explanation of spatial
scaling processes and their effects).

Syed et al. (
2005
)

used GRACE TWS estimates (which include changes in groundwater
storage

GWS) to estimate basin discharge, which they called ‘total basin dis
charge’ and
included the net of surface, groundwater and tidal inflows and/or outflows in addition to
streamflow. They found good correlation between streamflow and GRACE total basin flow,
although there were significant differences in magnitudes of low fl
ows (Amazon) and annual
amplitudes (Mississippi). They attributed at least part of these differences to changes in
GWSs.

Schmidt et al. (
2006
)

found that the hydrological signals of the world’s major river systems
were able to be recovered from GRACE data,

with a background model uncertainty of around
35 mm EWH from one month to another.

Crowley et al. (
2006
)

found significant seasonal variation and long
-
term loss of TWS in the
Congo Basin
. Syed et al. (
2008
)

found that GRACE
-
based
storage changes were in
good
agreement with those obtained from GLDAS simulations (e.g. 15 mm/month RMS between
the two estimates for the Mississippi River), whereas other authors have found significant
differences in ampl
itudes between GRACE and GLDAS (e.g. Tregoning et al.

2009
a
)
.

To put the TWS in perspective, the range of variation in TWS since the launch of GRACE in
2002 has been around ±300 mm in the Amazon Basin, while in the Murray

Darling Basin the
peak
-
to
-
peak changes are around 250

300 mm (
Leblanc et al. 2009
)
. Thus, a
potential
uncertainty of
approximately
30 mm represents around 10% of the anticipated changes in
TWS.

2.1.2 Extreme climate events: droughts and floods

Andersen et al. (
2005
)

identified a significant mass loss over Europe that occurred during a
record
-
brea
king heatwave in the summer of 2003. They estimated a loss of 78
±
10 mm EWH
from GRACE and confirmed this with GLDAS and a vertically integrated water balance
NATIONAL WATER COMMISSION


WATERLINES

7

estimate combined with a terrestrial water balance.
Chen et al. (
2009
)

provided quantitative
estim
ates of the extreme drought in the Amazon River
B
asin in 2005 using GRACE data. The
measurements were consistent with in situ water levels from river gauge stations and with
remotely sensed precipitation observations. However, they found that the land surf
ace models
significantly underestimated the intensity of the drought.

Reager and Famiglietti (
2009
)

used a combination of GRACE TWS and precipitation to derive
monthly storage deficit estimates and global maps of effective storage capacity from which
they
derived a monthly global flood index. Effectively, they identified cases where the
drainage systems were near capacity but precipitation continued
,

and used the information to
try to identify occasions of high likelihood of flooding events. The aim of this

work was to
present the information contained in GRACE data in a way that it may help to

predict

future
floods. Houborg et al. (
2010
)

also found that GRACE
-
based drought indicators contained
valuable information on drought conditions in addition to those
that rely heavily on
precipitation and do not account well for changes in SMS.

Steckler et al
.

(
2010
)

found an additional 50

Gt
6

of water storage in Bangladesh during
extreme flooding events, with GRACE estimates of the amount of floodwater agreeing withi
n
statistical limits with observed daily river levels. Chen et al.

(
2010
)

found peak flood flow
anomalies of 624
±
32 Gt for

the entire Amazon River
B
asin.


2.1.3 Groundwater studies

Rodell and Famiglietti (
2002
)

showed that it was feasible to use GRACE to s
ense
groundwater changes in the High Plains aquifer of the central USA, since the uncertainty of
the GRACE estimates was
around
8.7 mm compared
with

the observed periodic variations of
approximately
20

45 mm in GWS (note, however, that this does not includ
e any uncertainty
in soil moisture storage).
Post
-
launch studies found high correlations between GRACE TWS
and the sum of
GWS+SMS

(correlation coefficient

r
=
0.82) and GRACE and measured
groundwater variations (
r=
0.58) (Strassberg et al.

2
007
)
.


Yeh et al.
(
2006
)

found that groundwater estimates from GRACE agreed ‘reasonably well’
with in situ observations in Illinois, USA; however, they noted that the estimates differed
substantially in month
-
to
-
month variations. In general, the seasonal cycles between the
estimated and
measured

groundwater changes agreed well
(
r
=
0.83, 36 observations).
They
concluded that GRACE offered a means of estimating seasonal GWS changes at the basin
scale of 200,000 km
2
.

A similar study in the Mississippi River
B
asin found that it i
s possible to estimate variations in
TWS from GRACE, being the sum
of GWS, SMS and snow mass (Rodell et al.

2006
)
. This
study demonstrated how subtracting modelled estimates of snow and soil moisture (derived
from the GLDAS model) from the GRACE TWS estima
tes did yield groundwater estimates
that ‘compared favourably’ with well
-
based time series. However, the authors stated that the
results were better in basins larger than 900

000 km
2

than in sub
-
basins smaller than
500

000

km
2
. Thus, the relevance of GRACE

observations for smaller catchments remained
in question.

Leblan
c et al. (
2009
)

performed a study of the multi
-
year drought in the Murray

Darling Basin,
documenting the propagation of water deficits through the hydrological cycle. They found a
high corre
lation between the observed groundwater variations from boreholes and the
GRACE TWS estimates
,

at a time when the ongoing drought had reduced the available
surface water resources. The net loss of water over the period of GRACE observations
(2002

2007) was

found to be about 200 km
3
.
In
a similar study, Rodell et al. (
2009
)

quantified
the depletion of groundwater in India through a comparison of GRACE and GLDAS
observations as 109 km
3

over the period August 2002 to October 2008 (or 40

10 mm/year in
terms of
EWH). These two studies provided information averaged over
approximately
1

million km
2

and 450

000 km
2
, respectively.




6

1 gigatonne (Gt) of water is equivalent to 1 km
3

or 1000 GL

NATIONAL WATER COMMISSION


WATERLINES

8

Famiglietti et al. (
2011
)

followed a similar analysis approach to estimate that groundwater was
being depleted at a rate of 20.4±3.9

mm/y
ear (EWH) in the Central Valley, California,
amounting to around two
-
thirds of the total water loss. In this case, the basin has a size of
only
about
52

000 km
2
.

H
owever, the authors computed the GRACE TWS over the entire
Sacramento and San Joaquin basin r
egions (
about
154

000 km
2
) and then assumed that all
groundwater changes must have occurred only in the Central Valley region (since other parts
of the total region were mountainous and would have limited capacity to store gro
undwater).
Like Leblanc et al.

(
2009
)
, they found that groundwater depletion correlated with times of
drought.

More recently, Sun et al. (
2010
)

formulated a means of estimating aquifer storage parameters
from remotely sensed observations and modelled SMS estimates. They found that thei
r
estimated aquifer storage parameters were consistent with previous results derived from in
situ calibrations, and concluded that GRACE data can be used to derive spatially variable
parameters for groundwater modelling.

2.1.4 Validation of GRACE through g
round truth experiments

Because of the large spatial footprint of GRACE estimates of the
e
arth’s gravity field (around
380 km for a degree 50 spherical harmonic model
7
), it is extremely difficult to validate GRACE
estimates with in situ observations. Put s
imply, the spatial averaging that occurs when
generating a GRACE estimate of mass change is nearly impossible to replicate with discrete,
point
-
wise measurements. Nonetheless, several authors have found innovative ways in which
to validate the broad spatia
l estimates from GRACE using a range of different geophysical
signals.

Davis et al. (
2004
)

estimated
the pattern of annual deformation of the surface of the
e
arth
caused by annual variations in the global hydrological cycle. They compared the GRACE
-
derived

deformation
with

observed vertical surface movement at a GPS site at Brazilia in
South America and found very good agreement.
V
an Dam et al. (
2007
)

undertook a similar
study over Europe and concluded that there were significant differences between GRACE
a
nd GPS
-
derived deform
ations, while Tregoning et al. (
2009b
)

found very high correlations in
a similar comparison over the same region. The improved agreement in the latter study was
due to an improvement in the analysis of the GPS observations rather than
the identification
of any errors in the GRACE data.

Several authors have made comparisons of GRACE mass variation estimates and observed
ocean bottom pressure changes.

For example, Rietbroek et al. (
2006
)

found correlations of
0.7

0.8 between GRACE and oce
an bottom pressure observations in the Crozet
-
Kerguelen
region, while more recently Siegismund et al
.

(
2011
)

found globally averaged errors of 8.6,
11.1 and 5.7 mm EWH in a comparison of ocean bottom pressure variations and GRACE,
non
-
steric altimetry and
a climate/ocean model,
respectively. Tregoning et al. (
2008
)

compared sea surface height changes in the Gulf of Carpentaria estimated by GRACE with
tide gauge measurements and found excellent agreement in phase but small (<20%)
differences in amplitude. Th
ey also identified that the barotropic model
8

used in the reduction
of the raw GRACE observations underestimated the non
-
tidal ocean mass movement
significantly in the Gulf of Carpentaria
. Wouters and Chambers (
2010
)

reached a similar
conclusion from a stu
dy of ocean bottom pressure changes in the Gulf of Thailand, even
though the barotropic model in their analysis was not the same model.

Lo et al. (
2010
)

incorporated both GRACE TWS and estimated streamflow records to
constrain land surface model simulation
s and demonstrated the advantage of this coupled



7

Simplistically

speaking, the summation of many sine and cosine terms with different amplitudes and
periods allows complicated shapes and surfaces on a sphere to be represented by just the amplitudes
of the periodic terms. Thus, a representation of the Earth’s gravity fi
eld

either a mean field or at a
particular epoch

can be reduced to just a set of coefficients, known as Stoke’s coefficients, that are
multiplied by cosine and sine ter
ms. This is what is known as a '
spherical harmonic model
'
.

8

The barotropic ocean model,

described in Section 6.1.3, accounts for the gravitational effects on the
satellites from the non
-
tidal ocean mass movement.

NATIONAL WATER COMMISSION


WATERLINES

9

approach. They calibrated their model parameters using two years of data
,

then validated the
results using simulations spanning different time periods.

2.2 Review of applications of GRACE products
for hydro
logical studies in Australia

Despite the great technical success of the GRACE mission and the many different scientific
results that have been generated internationally, there are surprisingly few examples of the
use of GRACE data in Australia. Below, we d
ocument (in chronological order) the published
studies that we are aware of, including research driven by both national and international
scientists.

Rodell and Famiglietti (
1999
)

considered the Murray

Darling Basin in a prelaunch
assessment of what types
of hydrological signals would be detectable by the GRACE
mission. They concluded that monthly changes in TWS should be detectable over 80% of the
time, that the mean uncertainty would be 25

50% of the mean change in storage and that
seasonal and annual tre
nds would be detectable.

Ellett et al. (
2005a
)

presented the first assessment of the potential of the use of GRACE to
contribute to hydrological studies of the Murray

Darling Basin. They considered the
combination of GRACE with hydrological modelling, data

assimilation and ground
-
based
monitoring as a means of obtaining better resource management. Their initial results showed
the capability of GRACE to estimate statistically significant TWS changes on a basin scale
,

and the potential for these estimates to
improve model predictions in a data assimilation
framework. No actual GRACE results were presented (the data had not yet been made
publicly available); rather, the magnitudes of the hydrological signals were compared
with

the
expected errors of GRACE estim
ates based on prelaunch sim
ulation studies. Ellett et al.
(
2006
)

again proposed a framework by which GRACE observations could contribute to the
hydrological modelling of the Murray

Darling Basin but again did not use any GRACE data.

Ellett et al. (
2005b
)

p
rovided the first direct comparisons between actual GRACE estimates of
monthly TWS changes for the Murray

Darling Basin and those derived from two land surface
models and one rainfall
/
runoff model. They concluded from a comparison of data spanning
2002

200
4 that the differences were significant, with the models under
-

and over
-
predicting
the monthly mean water storages. This was the first use of GRACE data in a study of
Australian hydrology.

No further studies wer
e undertaken until Syed et al. (
2008
)

used G
RACE data (converted to
1º x 1º global EWH grids) spanning April 2002 to July 2004 to estimate a net depletion of 1.3
mm/month of TWS in Australia, with 1.1 mm/month of the total being lost from the

Murray

Darling Basin. This was the first quantification o
f water storage changes in Australia from
GRACE, albeit from only two years of data collected four years earlier.

Awange et al. (
2009
)

compared GRACE TWS estimates to rainfall data over Australia and
concluded that GRACE could detect hydrological signals.
However, they noted that the
relatively small hydrological signals over much of Australia were not detectable because of
errors in the GRACE data processing and the filtering methods that they had employed. They
indicated that an Australian
-
focused reproce
ssing of GRACE observations would be required
to reduce spectral leakage of ocean signals into continental estimates of TWS and to reach a
level of error smaller than the signals that are being sought.

Leblanc et al. (
2009
)

conducted a detailed study of th
e multi
-
year drought and its effect on the
Murray

Darling Basin. This was a comprehensive study that incorporated GRACE
observations, groundwater bore observations and estimates of surface water changes to
assess the response of water resources to the drou
ght and the assessment of its severity.
They found high correlations between TWS losses estimated by GRACE and depletion of
groundwater levels at a time when there was little change in modelled SMS and surface water
storage (the latter two had effectively
reached low values by the time the GRACE mission
NATIONAL WATER COMMISSION


WATERLINES

10

was launched or shortly after). This study showed, for the first time in an Australian context,
how GRACE data could provide important, basin
-
scale information on changes in TWS and
how, through integration
with soil moisture and surface water storage information,
groundwater variations could be sensed remotely.

Brown and Tregoning (
2010
)

investigated the magnitude of spectral leakage into estimates of
TWS in the Murray

Darling Basin from near and far
-
field
sources such as the Amazon Basin,
melting of Antarctica and Greenland and hydrological processes in Australia. They simulated
some of the world’s largest geophysical processes that have been detected by GRACE and
then assessed the amount of the simulated s
ignal that appeared in integrated TWS estimates
for the Murray

Darling Basin. The leaked signals into the basin reached maximum values of
approximately 10 mm EWH, which is around 30% of the formal uncertainty of GRACE
estimates and only about 10% of the ma
gnitude of changes in TWS that occur in the basin.

Leblanc et al. (
201
1
)

reduced the spatial extent to study groundwater changes in just the
Murray Basin (aproximately 300

000 km
2
, compared
with

approximately 1

000

000 km
2

for the
entire Murray

Darling Bas
in) and found a change in the long
-
term dynamics of the water table
since the onset of the drought in 1997. Borehole data showed a regional increase in the water
table from 1980

1992, then a steady decline (around 17 cm/year) from 1997 to 2009. Over
the GR
ACE period, groundwater losses of 18±1.3 mm/year have occurred (derived from
GRACE TWS minus modelled soil moisture values), equating to about 45±3 km
3

integrated
over the basin. They argued that the drought (temporarily) reversed the impacts of past land
clearing in creating dry land and salinity problems.

Awange et al. (
2011
)

investigated the use of
4°×4° resolution '
mascon
'

(mass concentration)
GRACE solutions (see Section 3.1.5 for more details) over Australia for monitoring
hydrological processes. They

extracted from the mascon solutions the main spatial and
temporal components (rate, annual trend
,

etc
.
) but concluded that, when considering Australia
as a whole, the mascon approach (at least, at the 4°×4° resolution) did not contribute
significantly mor
e information than the available spherical harmonic solutions.

Frappart et al.
(
2011
)

developed a series of solutions using an Independent Component
Analysis for the Murray

Darling Basin. They found that their solutions agreed better with the
in situ obser
vations than the other spherical harmonic solutions that had undergone various
types of filtering and rescaling (see Section 3), with the maximum deviations between GRACE
and in situ observations decreasing by a factor of two to three. This shows the poten
tial to
improve the accuracy of GRACE estimates through more appropriate statistical treatment of
the data.

García
-
García et al. (
2011
)

analysed GRACE data from 2002 to 2010 and found that 60% of
the variance across the Australian continent could be accoun
ted for with an annual periodic
signal. They found that phases of the Indian Ocean Dipole (IOD) were correlated with
precipitation in south
-
eastern Australia associated with changes in tropical moisture flux. They
noted, in particular, that the dry period
of 2006

2008 coincided with three consecutive periods
of positive IOD events.

Va
n Dijk et al. (
2011
)

compared the AWRA hydrological/land surface model with GRACE
estimates of TWS across the Australian continent. This is the most extensive comparison of
GRA
CE and hydrological models over Australia and is discussed further in Chapter 4.

2.1.4 Summary

The results and conclusions of the above studies demonstrate clearly the potential of GRACE
to contribute significant and unique information regarding changes in

total water storage over
the Australian continent. While it has not been
,

and may never be
,

demonstrated that
estimates can be made at spatial scales as small as individual farms or basin sub
-
catchments, the ability to provide over
-
arching constraints on
the total water storage at the
200

000 km
2

scale is feasible. Researchers have already shown how such information can be
used to study aspects of hydrology as diverse as the severity of droughts, constraining
NATIONAL WATER COMMISSION


WATERLINES

11

specific yield values, estimating groundwater s
torage changes, guiding the development and
improvements in hydrological models and even identifying surface deformation caused by
hydrological loading.

NATIONAL WATER COMMISSION


WATERLINES

12

3. Assessment of the available
GRACE gravity fields

Several international centres use the original GRA
CE satellite observations to derive temporal
estimates of the
e
arth’s gravity field and provide these as products in the form of spherical
harmonic coefficients (defined below). The available products have different time intervals
(daily, 10
-
day and 30
-
day

averages) and are generated using a range of different analysis
strategies. Consequently, the way in which the gravity fields generated by different
international groups should be used is also different. Incorrect use can result in wildly
incorrect estima
tes of hydrological variables. Additionally, some centres now provide global
grids of estimates of changes in mass in terms of EWH. The suite of available information can
be confusing for users not familiar with the technical details of the analysis proces
ses.

To date, no comprehensive assessment of these different solutions has been made and it is
not well known to what extent the hydrological estimates across Australia would differ
between solutions; however, a preliminary analysis of only three of the av
ailable products
showed considerable dif
ferences in quality (Van Dijk et al.

2011
)
. Below we provide a short
description of the analysis strategies of several available GRACE gravity fields, as well as
brief explanations of how each centre indicates that t
heir GRACE products should be used. In
this report we use products provided by the French interagency Space Geodesy Research
Group (
Groupe de Recherche de Géodésie Spatiale
, GRGS)

and
the Center for Space
Research of the University of Texas at Austin (
CSR
)
.
We discuss
methods recommended for
reducing correlations between parameters of the spherical harmonic coefficients, reducing
leakage of ocean and land signals into other regions, application of spatial filtering to mitigate
high levels of noise in the hi
gher degree spherical harmonic coefficients and, subsequently,
the rescaling of resulting solutions to mitigate the loss of signal from the filtering processes.

3.1 GRACE products and their use

The shape of the
e
arth is commonly referenced to its gravitati
onal equipotential surface called
the geoid. The geoid is a useful reference since it is the surface that the
e
arth’s sea level
would describe in the absence of winds, ocean currents, and other non
-
self gravitational
disturbing forces. The geoid provides a
ccess to the local up/down direction and the horizontal
plane. In mathematical models, the
e
arth's first order shape is conveniently described as an
ellipsoid, where the equatorial radius is about 21 km greater than the polar radius. Departures
of the
e
art
h’s topographic relief, and geoid, are represented as elevation above or below its
best
-
fitting reference ellipsoid. The
e
arth’s geoid is up to 110 m below and 90 m above the
reference ellipsoid, while its topographic surface can be up to 11

000 m below an
d
approximately 9000 m above this reference ellipsoid.

The
e
arth's gravity field is determined by how the material that makes up the
e
arth is
distributed. Because gravity changes over the surface of the
e
arth, the weight of an object
changes along with it.

For convenience we represent the
e
arth’s gravity field as the sum of a
smooth standard
e
arth gravity model (Figure 2a), and gravity 'anomalies' (Figure 2b) which
describe how actual gravity deviates from the standard model. A map of gravity anomalies
(usu
ally expressed in units of milliGals
9
) tends to highlight short wavelength features better
than a map of the full geoid.

Historically, geodetic analysts have produced representations of the
e
arth’s gravity field using
spherical harmonic models. These have
been derived since the 1970s from the observations
of the motion of satellites orbiting the
e
arth, with a trend of gradual increases in accuracy as
more observations became available. A quantum leap occurred with the GRACE mission
because, for the first ti
me, inter
-
satellite range changes could be used to ma
p changes in the
gravity field (Tapley et al.

2004
)
. Today, gravity field estimates are available for both the mean



9

A
G
al, short for Galileo
,

is a unit of measure of acceleration and is equal to 0.01

m/s².

NATIONAL WATER COMMISSION


WATERLINES

13

(or static) field and for means of particular time interval
s of 1, 10 or 30 days durati
on (e.g.
Tapley et al.

2004; Kurten
bach et al. 2009; Bruinsma et al.

2010
)
.

Figure
2
: The
e
arth’s
gravity

field, showing a) the latitudinal variation

caused by the equatorial
bulge
,

b) a snapshot of geophysical processes by comput
ing anomalies at a single epoch (i.e.
residual signal about the mean value)


The original approach of the GRACE
s
cience
t
eam was to develop spherical harmonic
models for 30
-
day epochs from the GRACE observations, and solutions by CSR, the
German
Research
Centre for Geosciences
(GFZ) and Jet Propulsion Laboratory (JPL) are available.
Essentially the observations (the satellites’ positions/velocities and the changes in the inter
-
satellite distance) are related to the parameters (the Stoke’s coefficients)
,

an
d a linear
inversion yields estimates of the spherical harmonic model(s) of the gravity field(s).
Subsequently,
t
he French GRGS developed spherical harmonic models as did the Institute of
Geodesy and Geoinformation, University of Bonn (ITG). Differences be
tween the approaches
used to generate the models mean that the solutions are not exactly the same, as will be
explained below.

An alternate approach has been used to localise the changes in the gravity field into regions,
then estimate mass changes for ea
ch region (assuming a constant mass change across each
region). This
'
mascon
'

approach was developed for studies of Venus and was first applied to
the analysis of

GRACE data by Rowlands et al. (
2005
). Luthcke et al. (
2006
)

used a similar
approach to study
mass balance changes of Greenland and

g
lobal mascon solutions of a 4º x
4º degree grid are now pub
licly available. Awange et al. (
2011
)

assessed the feasibility of
using these grids for studying hydrological processes in Australia and found no significant
improvement over using the more conventional spherical harmonic fields.

3.1.1 Underlying model assumptions

The process of estimating mass changes on
e
arth from the original GRACE observations is
complicated and involves many detailed steps. The motion of t
he satellites is governed by the
shape of the
e
arth’s gravity field as well as the gravitational attractions of the
s
un,
m
oon and
other planetary bodies, although the
e
arth’s gravity field exerts the greatest force, since it is
the closest to the satellite
s. It is comprised of many different components:




the static (or constant) gravitational field caused by the mass of the
e
arth (known as the
central body force)



the change in gravity caused by the deformation of the solid
e
arth as a result of the
gravitati
onal forces of the
s
un and the
m
oon. This is often called the '
solid
e
arth
'

or
'
body
'