Content uploaded by Andrea Gatti
Author content
All content in this area was uploaded by Andrea Gatti on Apr 12, 2017
Content may be subject to copyright.
Gatti, A.; Reguzzoni, M.; Migliaccio, F.; Sansò, F. Politecnico di Milano,
Italy
Computation and assessment of the fifth
release of the GOCE-only space-wise solution
!The fifth release of the space-wise approach, as well as the ones of the direct and time-wise approaches, is based on the processing of the whole GOCE dataset, from
November 2009 to October 2013. It consists in global grids of gravity gradients at 0.2°x0.2° spatial resolution and in a spherical harmonic model derived from these grids by a
discretized quadrature formula and by a proper global regularization. Output products will be distributed through the ESA website and the ICGEM service
!The main difference between the space-wise approach and the other two official methods is that the GOCE data acquired along the orbit are compressed onto global grids by
applying collocation on local data patches, each of them with its own locally adapted signal covariance. In this way the filtering level is locally controlled, differently from the
other solutions where a global regularization on spherical harmonic coefficients is applied
!The good quality of the space-wise products has been assessed by comparing them with other grid solutions (e.g. the ones computed in the framework of the STSE GOCE+
GeoExplore project) and with other spherical harmonic models (e.g. the ones by the direct and time-wise approaches). The latter comparison also indicates the weakness of
the local collocation method to optimally estimate the low degrees of the gravity field spectrum but, at the same time, its better capability to recover the highest degrees where
the regularization strategy plays an important role
ABSTRACT
Space-wise products - release 5
Pre-Processing
GRIDSSH model
LS SST SST local
gridding
Wiener filter
+ local gridding
Whitening filter
+ local gridding
analysis and global regularization
The pre-processing is a fundamental step to be performed before using GOCE
data. Each dataset (positions,
accelerations, rotations and gradients)
used for the solution is checked,
outliers are removed, and whenever
possible filled with reasonable
interpolated data that help the filtering
step
PRE PROCESSING
09/2009
01/2010
04/2010
07/2010
10/2010
01/2011
04/2011
07/2011
10/2011
01/2012
04/2012
07/2012
10/2012
01/2013
05/2013
08/2013
0
10
20
30
XX
XZ
YY
ZZ
% of flagged data
per gradient
(map of XY not
shown here)
(below) % of flagged data per gradient per cycle
SST SOLUTION
The SST solution is computed in two steps, a Least Squares adjustment up to
degree 140 considering only error variances of the along orbit potential,
followed by a spherical gridding @229km based on local signal covariances
the SST gridding is masked to
provide corrections only where
the SNR is high enough
[m2/s2]
[m2/s2]
SGG FILTERING
The gradients are prepared for the gridding steps by removing a low frequency
systematic effect in ascending and descending orbits. Wiener and White filters
are almost complementary.
MF grid
weights
HF grid
weights
Collocation weights (from Trr to Trr)
for the two gridding steps:
Medium Frequencies and High Frequencies
Frequency
10-6 10-4 10-2
0
0.2
0.4
0.6
0.8
1Wiener filter on Trr
Whitening filter on Trr
Spectral response of the
Wiener and Whitening filter
[Hz]
SGG GRIDDINNG
In the first gridding step, applied to the Wiener filtered signal, grids of different
derivatives w.r.t. r and lambda are estimated. By analysing these grids we derive
different sets of spherical harmonic coefficients that are combined to produce a
correction to the SST solution. The combination is driven by Monte Carlo (MC)
samples that pass through the same processing of the data.
In the second gridding step, after the Whitening filtering, grids of the full tensor
are estimated
,
Here we show the differences of the solution
w.r.t. synthesized spherical grids of Trr having
the same radius of 6600 km (229 Km altitude).
As a comparison we show the results using
the DIR R5 and TIM R5 models too
Differences between the two GOCE-only models (TIM and SPW)
can be seen at low frequencies and at high frequencies in areas
where a strong signal is present
Empirical Difference Degree Variances reveal the good performance of the space-wise solution but
for the low degrees below 80. Coefficients directly coming from the grids (black) are regularised
according to MC samples to produce a set of regularised coefficients (green)
The power of the space-wise coefficients
derived from the GOCE grids, even when
regularised, is higher than the one of the other
official GOCE solutions, this is probably due to
the type of local covariance adaptation
implemented in the gridding
disregarding low
order coefficients
considering low
order coefficients
Synthesis of Trr grids of the differences between
the SPW coefficients and TIM release 5 solution
above degree 230 (right)
Space-wise and STSE GOCE+ solutions over an
ellipsoidal grid at 225 km are finally compared with
a grid computed from GOCO up to degree 1440,
thus including very high frequency information
(see below for results)
•Spherical grids of gravity gradients by the space-wise approach have been computed and will be distributed by ESA
•From the grids, a set of spherical harmonic coefficients is derived and compared with other models, showing good performance at high degrees. This model will be made
available through the ICGEM website after a final refinement/optimization
•In the future, besides optimizing the present solution, a new model based on GRACE prior information will be computed, to be compared e.g. with DIR and GOCO solutions.
Local grids will be also studied for the ISG service
Conclusion
degree
050 100 150 200
degree variances [unitless] Log scale
10-20
10-18
10-16
Error (Difference) Degree Variances
DV EIGEN 6C4
EDV LS SST
EDV LS gridding
estimated error
disregarding low
order coefficients
degree
050 100 150 200
degree variances [unitless] Log scale
10-20
10-18
10-16
Error (Difference) Degree Variances
DV EIGEN 6C4
EDV LS SST
EDV LS gridding
estimated error
considering low
order coefficients
[mE]
[mE]
[mE]
[mE]
[mE]
[mE]
[mE]
[mE]