Professional Documents
Culture Documents
Duan 2012.spurious Jump RL05
Duan 2012.spurious Jump RL05
EXPRESS LETTER
Accepted 2012 August 1. Received 2012 July 31; in original form 2012 June 5
C 2012 The Authors 83
Geophysical Journal International
C 2012 RAS
84 J. Duan et al.
melting, consequently contaminating regional mass transport esti- 2006 and 2010. To be complete, we have also included in Fig. 1 the
mates from GRACE data. These jumps may attain 7 cm of equiva- changes computed from GAA data, which include the contribution
lent water thickness (EWT) change in certain regions, for example, of the atmosphere alone. We see that in both GAC and GAA data,
in Qinghai-Tibetan Plateau and the western part of South America the short wavelength jumps are present over land. While zooming to
with voluminous ice coverage. We use South America as an example very large scale maps, we see that the largest jumps seem to occur
to quantitatively illustrate how the glacier mass balance estimates at locations with steep terrain elevation changes, for example, the
using GRACE L2 data could be contaminated by the jumps. Tibetan Plateau and Tianshan in China, the Andes in South America
These jumps are likely caused by the horizontal and/or ver- and the Rocky Mountains in North America. Ripples over the oceans
tical resolution change in the European Center for Medium- are more likely caused by spherical harmonics truncation used to
range Weather Forecasts (ECMWF) operational atmospheric model represent GAA/GAC.
adopted for deriving the AOD1B data. We find that the jumps oc- These jumps can be seen much more clearly by zooming to a
curred on 2006 February 1 and 2010 January 26, respectively (http:// smaller region. Here we illustrate them using an Rotated Empiri-
www.ecmwf.int/products/data/operational_system/evolution/). cal Orthogonal Function (REOF) (Richman 1986) analysis on the
monthly GAC solutions for the Tibetan Plateau. The EOF analysis
decompose the original monthly solutions into a number of EOFs,
2 S Y S T E M AT I C B I A S E S I N T H E
which are functions of locations. Each EOF represents a spatial
A O D 1 B D ATA
pattern, while an accompanying time-series called principle com-
We examined both the Release 4 (RL04) and 5 (RL05) of the ponent (PC) represents the temporal dependence of the pattern. The
Level-2 data (GAA and GAC products) applied by all three offi- product of the EOF and the PC gives the values of the pattern as
cial GRACE data product centres: the University of Texas at Austin function of location and time in real physical dimension. One of the
Center for Space Research (CSR), the German Research Center for EOF/PC pair depicts each jump. The two jumps have similar but still
Geosciences (GFZ), and the Jet Propulsion Laboratory (JPL). Since distinct spatial patterns. Hence, we use two different fractions of the
all these three centres are using the same AOD1B product delivered 9 yr data to catch the two jumps individually, one from 2003 January
by GFZ, the differences among them are insignificant for our pur- to 2008 December, the other from 2007 January to 2011 Decem-
pose. The difference between RL04 and RL05 AOD1B products is ber, respectively. The leading and second EOFs/PCs describe the
that an updated ocean model is adopted for RL05 products. Consid- seasonal and interannual signals, which are not within the scope of
ering that RL04 data are available for longer time span than RL05 this study, thus are not shown here. Fig. 2 shows the third EOF/PC
data at the moment, and the same atmospheric data are used in both which represents the jump. The EOF patterns are normalized by
RL04 and RL05 products. We choose to present our results based setting the absolute maximum to 1, so that the PC time-series rep-
on CSR AOD1B L2 GAC and GAA RL04 data for 9 yr spanning resents the largest changes with the dimension of EWT change in
2003/01-2011/12, and focus mainly on the continental area. millimetre. The EOF patterns for the two time spans spatially match
The L2 GAC data are the monthly means of the geopotential the year-to-year changes shown in Fig. 1, respectively, and the time-
due to variations in the atmosphere and oceans computed from six series provide a clear view of the spurious jumps from January to
hourly AOD1B atmospheric and oceanic models, which have the February in 2006 and 2010, respectively. From the figure of the PCs
same time span as the L2 monthly geopotential solutions, while the we see the jumps attain a magnitude of 7–8 cm.
GAA data represent the geopotential variations due to atmosphere Since we could not find any evidence to support the hypoth-
only (Flechtner 2007). Each monthly GAC/GAA solution consists esis that this kind of terrain-related jumps are real climate sig-
of a set of spherical harmonic (Stokes) coefficients from degree 0 nals, we argue that they are most likely systematic biases in the
up to a maximum degree 100, but only the coefficients from degree atmospheric model adopted. Because surface pressure is the most
2 onwards are used. important parameter in the computation of GAA/GAC products,
We first present a global view of the jumps, although the jumps if the jumps are real signals, they should be reflected in changes
could be more pronounced regionally such as at or around the of surface pressure values. To test our hypothesis, we have also
Tibetan Plateau, which will be presented below. We use the changes used the reanalysed monthly surface pressure data of the ECMWF
of the averages of two consecutive years of the GAC solutions to Re-Analysis (ERA-interim) products (http://ecmwf.int/research/
more definitively detect the jumps, since signal of seasonal changes era/Project/Plan, 2000), which are generated with consistent res-
is usually diminished in the yearly average. As both jumps occurred olutions, to show the corresponding yearly changes as shown in
from January to February, we compute the yearly averages using Fig. 1 for comparison (Fig. S1, Supporting Information). The ERA-
monthly solutions from February to January in next year, so that no interim surface pressure data are converted into surface load in
jumps due to real signals, that is, seasonal variations, are included terms of height of water column by following Flechtner (2007). All
in the yearly averages. We roughly call the average from 2003 figures are similar to the top panel of Fig. 1. The short wavelength
February to 2004 January the yearly average of 2003 and so forth. features did not show up in the ERA-interim data, thus supports the
The jumps are presented in Fig. 1 using EWT computed according hypothesis that the jumps in the GAC data are spurious. Actually,
to Wahr et al. (1998). We first present the change from 2006 to the jumps are very likely related to the changes of resolution in the
2007 (top panel), which is considered as the ‘normal interannual atmospheric model, which will be further discussed later.
change’. The changes from 2003 to 2004, from 2004 to 2005, from We have chosen the Tibetan Plateau, since the jumps are rela-
2009 to 2010 and from 2010 to 2011 are all similar. We then present tively larger over there, and that the quantification of the mountain
the changes from 2005 to 2006 (middle panel) and from 2009 to glacier mass balance over there remains one of the major research
2010 (bottom panel) when the jump happened. As compared to the interests (e.g. Matsuo & Heki 2010; Jacob et al. 2012). It can be
‘normal interannual change’ on the top panel, the middle and bottom readily inferred from Fig. 2 how much glacier mass balance could be
panels include substantial spatially short wavelength changes. These contaminated by the spurious jumps, as these spurious jumps would
short wavelength changes are likely caused by the jumps in the produce corresponding jumps in mass changes with the same mag-
AOD1B data, which happened from January to February in both nitudes but with opposite signs.
Figure 1. Changes of yearly averages of surface mass anomaly (in terms of equivalent water thickness in millimetre) showing a global view of the jumps in
the de-aliasing product from January to February in both 2006 and 2010. Left panel presents the results of GAA data. Right panel presents the results of GAC
data. From top to bottom are the changes from 2006 to 2007, from 2005 to 2006 and from 2009 to 2010, respectively. The top panels represent normal change
from 1 yr to the next one without jump in between the 2 yr. As compared to the normal change shown in the top panels, the changes with a jump in between
the 2 yr show a lot of short wavelength spatial features, which are the jumps. All changes not shown are similar to the top cases, that is, without the short
wavelength features.
C 2012 The Authors, GJI, 191, 83–87
Geophysical Journal International
C 2012 RAS
86 J. Duan et al.
Figure 2. (a) Spatial pattern of the third EOF over the Tibetan Plateau during 2003/01–2007/12; (b) its PC time-series; (c) spatial pattern of the third EOF
over the Tibetan Plateau during 2007/01–2011/12; (d) its PC time-series.
Figure 3. (a) Ice-covered regions in South America, from the fig. 1 in Jacob et al. (2012) (Courtesy Dr. John Wahr); (b) year-to-year change of GAC surface
mass anomaly between 2006 and 2005 over South America; (c) year-to-year change of GAC surface mass anomaly between 2010 and 2009 over South America;
(d) mass changes during 2003–2011 for ice-covered region 16 in (a), blue line shows mass change from monthly GSM solutions, red line shows surface mass
anomaly from monthly GAC data with sign changed (multiplied by (−1)).
mass balance estimate is actually a distortion due to the spurious project within the Ohio State University’s Climate, Water, and Car-
jump in AOD1B data. The drop in glacier mass balance estimate is bon (CWC) Program, and Chinese Academy of Sciences/SAFEA
in agreement with the result of Jacob et al. (2012) during 2003–2010 International Partnership Program for Creative Research Teams
for this region, which show a sudden drop in early 2010. (Grant No. KZZD-EW-TZ-05). We thank Frank Flechtner and
Thomas Gruber for their constructive comments, which resulted
in an improved manuscript. Some of the figures in this paper are
4 C O N C LU D I N G R E M A R K S generated using the Generic Mapping Tool (GMT) (Wessel & Smith
We find two large jumps with magnitudes attaining about 7 cm of 1991).
EWT change in the AOD1B atmospheric pressure variation data for
both the RL04 and RL05 AOD1B data products from January to
February in both 2006 and 2010, respectively. REFERENCES
A global view of the jumps are shown using the changes of the
two yearly averages of atmospheric pressure just before and after the Chambers, D.P. et al., 2004. Preliminary obsrevations of global ocean mass
jumps in contrast to a case without the jumps, as seasonal changes with GRACE, Geophys. Res. Lett., 31, doi: 10.1029/2004GL020461.
are usually reduced in the yearly averages. The jumps have a short Chen, J.L. et al., 2006. Satellite gravity measurements confirm accelerated
melting of greenland ice sheet, Science, 313, 1958–1960.
wavelength spatial feature, which is demonstrated in more detail for
Flechtner, F., 2007. AOD1B product description document for product re-
the Tibetan Plateau based on an EOF analysis. They are likely largest
leases 01 to 04, GRACE 327-750, GFZ publ. GR-GFZ-AOD-0001 Rev.
in regions with steep changes in terrain altitude over a short distance. 3.1, GeoForschungsZentrum Potsdam, Germany.
As we could not find any causes in atmospheric processes which Han, S.-C. et al., 2006. Crustal dilatation observed by GRACE after the
could cause these jumps, we argue that they are spurious facts in the 2004 Sumatra-Andaman earthquake, Science, 313, 658–662.
atmospheric data product. In support of our conjecture, no jumps Jacob, T. et al., 2012. Recent contributions of glaciers and ice caps to sea
are found in the corresponding times in the reanalysed monthly level rise, Nature, 482(7386), 514–518.
ERA-interim atmospheric pressure data. Hence, the jumps appear Matsuo, K. & Heki, K., 2010. Time-variable ice loss in Asia high mountains
only in the ECMWF operational model adopted for AOD1B. We from satellite gravimetry, Earth planet Sci. Lett., 290, 30–36.
find that the spurious jumps are likely related to resolution changes Morison, J. et al., 2012. Changing Arctic ocean freshwater pathways mea-
sured with ICESat and GRACE, Nature, 481, 66–70.
in the atmospheric model. On 2006 February 1, there is an increase
Richman, M., 1986. Rotation of principal components, Int. J. Climatol., 06,
of both vertical and horizontal resolutions, and on 2010 January 26,
293–335.
there is an increase of horizontal resolution. In both cases, when Rodell, M. et al., 2009. Satellite-based estimates of groundwater depletion
the jumps occurred, the horizontal resolution of the operational in India, Nature, 460, 999–1002.
atmospheric model was increased, from T511 to T799 (from 40 to Swenson, S.C. & Milly, P.C.D., 2006. Climate model biases in seasonality of
25 km) in 2006 and from T799 to T1279 (from 25 to 16 km) in 2010. continental water storage revealed by satellite gravimetry, Water Resour.
This implies that the orography used in the atmospheric model was Res., 42, doi: 10.1029/2005WR004628.
increased accordingly. Surface pressure, which is the most important Tapley, B.D. et al., 2004a. The gravity recovery and climate experi-
parameter in the computation of the AOD1B products, changes ment: mission overview and early results, Geophys. Res. Lett., 31, doi:
with height. Therefore, with increased orography resolution, surface 10.1029/2004GL019920.
Tapley, B.D. et al., 2004b. GRACE measurements of mass variability in the
pressure values would be changed, especially in such areas with high
earth system, Science, 305, 503–505.
terrain variability.
Velicogna, I. & Wahr, J., 2006. Measurements of time-variable gravity show
As the spurious jumps in the atmospheric de-aliasing product mass loss in antarctica, Science, 311, 1754–1756.
cause uncertainties in the monthly GRACE geopotential solutions, Wahr, J. et al., 1998. Time variability of the earth’s gravity field: hydrological
which in turn cause uncertainties in mass transfer estimates on and ocean effects and their possible detection using GRACE, J. geophys.
Earth, for example, for mountain glacier mass balance. Therefore, Res., 103, 30 205–30 229.
quantifying and correcting the impacts of the systematic biases in Wessel, P. & Smith, W. H. F., 1991. Free software helps map and display
the GRACE de-aliasing products is critical. We propose a post- data, EOS, Trans. Am. geophys. Un., 72, 441, 445–446.
processing correction approach of the jumps for regional mass bal- Zenner, L. et al., 2010. Propagation of atmospheric model errors to grav-
ance studies based on an EOF analysis. Take for example, the jump ity potential harmonics-impact on GRACE de aliasing, Geophys. J. Int.,
182(2), 797–807.
from 2006 January to February in Tibetan Plateau shown in Fig. 2.
Zenner, L. et al., 2012. Non-tidal atmospheric and oceanic mass variations
We first compute the averages of PCs during 2 yr before and after
and their impact on GRACE analysis, J. Geodyn., 59-60, 9–15.
the jump, respectively. We then subtract the average before the jump
from the average after the jump. Finally we multiply the difference
with the EOF to obtain the jump as function of location, which is
S U P P O RT I N G I N F O R M AT I O N
referred to as true signal jump (TSJ). To remove the effect of the
jump from GSM data, the TSJ should be added to all monthly GSM Additional Supporting Information may be found in the online
data after the jump. version of this article:
Finally, we suggest that beside this artificial and cumbersome
correction approach the next release of AOD1B (RL06) should Figure S1. Year-to-year changes of annually averaged surface mass
be based on ERA-interim products, or a separate pre-processing anomaly from the ERA-interim surface pressure product, from top
step for ECMWF atmospheric model data should be applied during to bottom are the changes from 2006 to 2007, from 2005 to 2006
GRACE data processing in case such jumps occur. and from 2009 to 2010, respectively.
C 2012 The Authors, GJI, 191, 83–87
Geophysical Journal International
C 2012 RAS